Sophia and her critics
The ethics of human likeness — Part 1
The CogX conference last week hosted a debate on the question, “Should robots resemble humans?” Representing the “No” response was Joanna Bryson and Alan Winfield, both academics specializing in robot ethics. Representing a qualified “Yes” response was David Hanson and Will Jackson, both roboticists directing private research into human-like robots. Kate Devlin was a wonderful moderator. CogX deserves recognition for hosting the event. Although short, I think the debate was illuminating. It revealed wide common ground and very little disagreement on either side.
Someone who isn’t already immersed in the debate might find that rather complicated issues go by quickly, and they might wonder where the disagreement actually lies. In this essay I hope to convey a sense of why this event deserves broader recognition and discussion, especially in the AI & ethics community. In Part 1 I’ll provide some background for this debate in the form of an extensive timeline and analysis of Sophia’s media performances and the rise of criticisms against her. In Part 2 I’ll discuss Joanna Bryson’s views on robots and personhood in more detail. Then I’ll discuss the CogX debate itself and its aftermath, and offer my own thoughts on the ethics of human likeness. I’ll close the essay by considering sex robots as a particularly vexing case where human likeness is a central concern.
Like the representatives of the “no” side, I am also an academic who works on robot ethics. I’ve been critically engaged with participants from both sides of the debate on social media over the last year as the controversies around Sophia have swelled. As I’ll detail below, I find both positions in this debate frustrating. I think both sides fail to appreciate a critical factor at play in the ethics of social robots — namely, how the construction and use of social categories impact the dynamics and norms of social organization. I’d like to use this debate to draw attention to the way we’re thinking about the category of “robot”, and the influence these discussions have over the development of our ethical practices towards them.
To be clear, it is not my intention to undermine the respectful, engaging conversation on stage, or to undo whatever good will has been generated by this event. Although I disagree with both sides on certain issues, and although the disagreement concerns ethical matters, my comments here are offered to continue in the same earnest, collegial, professional tone of the debate’s participants, all of whom have contributed an important perspective to the conversation and have earned their position on stage. I offer my commentary out of a sincere concern for the ethics of AI, a concern I know I share with everyone involved.
I‘ll begin with a timeline of Sophia and her critics. A full appreciation of this background will help cast all sides on the debate in their best light possible, including my own views and criticisms, which I will try to save until the end.
A pre-debate timeline
At the start of the debate, Kate mentions a “Twitter pre-debate”, but the controversy motivating the debate goes back at least a full year and deserves careful documentation. The controversy centers on Sophia, the world-famous robot built by David’s company, Hanson Robotics. While wildly popular in the media, Sophia has received strong criticism from the AI community. Joanna called Sophia “obvious bullshit” last October, a sentiment upgraded in January to “complete bullshit” by Yann LeCun, then director of Facebook AI Research. As far as I’m aware, this CogX debate is the first public event where Sophia’s representatives have responded directly to criticisms from the AI and ethics community.
A conflict immediately preceding the Twitter pre-debate gives some insight into how tensions around Sophia have grown over the last year. CogX originally listed Sophia as an invited speaker, potentially including her in the conversation with Joanna and David. Ads promoting Sophia and Joanna started going out a month before the event.
Joanna was asked to comment on Twitter about being listed alongside Sophia for the event. Joanna’s response was critical, putting Sophia “on par with a scam.”
Although the organizers addressed Joanna’s concerns quickly, the handling of Sophia’s presence continued to stir controversy. A week later, hal hodson called CogX a “garbage merchant” over the promotion of Sophia, again triggering responses from Bryson and the CogX organizers.
This minor clash over whether and how to include Sophia, and her frequent association with bullshit, garbage, and scams, helps explain Kate’s introduction of red and yellow cards at the start of the debate, and David’s emphasis on “civility” and caution against “prohibition and name calling” during his opening remarks (6:08), comments which felt incongruous with the respectful, professional discussion that actually occurred.
The following timeline is not comprehensive, but is meant to sketch Sophia’s arc through the media circus, and to highlight some important criticisms and responses prior to the debate. I’ve given particular detail around October and November of 2017 when many key incidents take place, including the announcement of the SingularityNet ICO, Sophia’s speech at the UN, the announcement of Saudi citizenship, and the first rumblings of criticism against her.
I should also clarify that there are several distinct instances of Sophia floating around (by some counts twelve). Some are more sophisticated than others. This explains why Sophia can sometimes appear in two places at once, such as being both in Australia and Saudi Arabia for separate events on October 25th. When not with David Hanson, Sophia most often appears in public with the chief scientist at Hanson Robotics, Ben Goertzel.
- 2015 — Sophia is first turned on in April. She is described as “the most beautiful and celebrated robot” on hansonrobotics.com. Source code for Sophia’s core software platform, OpenCog, is freely available on GitHub by the end of the year, as are software packages for other Hanson Robotics systems.
- March 2016 — David and Sophia appear at SXSW and give an interview with CNBC that goes viral. Sophia’s first major introduction to mainstream western media describes her as a “hot robot” that “wants to destroy humans”. Note that David comments directly on the issue of deception and human likeness in this interview. He claims a preference for robots to “look a little like a robot, so you know.” Of course, Sophia’s head includes a plastic, transparent panel that reveals the animatronics driving Sophia’s facial expressions. In the clip, David explains his dream of building robots that are “genuinely alive”, and he imagines what robots will be capable of in the future. Hanson’s comments reads in the clip like an implicit acknowledgment of Sophia’s limitations. Still, David emphasizes that that his robots are “extremely lifelike” and capable of natural language understanding and learning. There is no disclosure that some of Sophia’s dialog for the interview is scripted by her handlers.
- October 2016 — Charlie Rose for 60 minutes interviews David and Sophia. Rose questions Sophia about whether she is “spontaneous” during what appears to be an unscripted, interactive chat session.
- November 2016 — Sophia and Ben appear at the Web Summit 2016 in Libson. At Clockenflap 2016 in Hong Kong, Sohpia performs All is Full of Love. The official Hanson Robotics YouTube channel publishes “Sophia Awakens Episode 1 and 2”, dramatizing an encounter between Sophia and one of her creators. Sophia’s voice actor and writer are given explicit credit in these short films.
- January 2017 — The European Union passes a draft report recommending various laws and policies concerning robotics. The report includes a call (31) to “explore the implications of all possible legal solutions” to a number of problems with robotics, including (f) “creating a specific legal status for robots… with certain rights and obligations”. The media widely report the story as a call to give robots personhood status in the EU. It is worth noting that Sophia is rarely mentioned in connection with this story.
(31f) — creating a specific legal status for robots, so that at least the most sophisticated autonomous robots could be established as having the status of electronic persons with specific rights and obligations, including that of making good any damage they may cause, and applying electronic personality to cases where robots make smart autonomous decisions or otherwise interact with third parties independently.
- April 2017 — Sophia appears on Jimmy Fallon. Fallon suggests (and Hanson agrees) that Sophia is “basically alive”. Sophia’s material appears entirely scripted, though this is not made explicit during the performance.
- June 2017 — Sophia opens CogX 2017 in London. Hanson appears on a panel where he announces plans for Sophia to become “genuinely alive” in the next 5–10 years. Subsequently, Piers Morgan interviews Sophia on Good Morning Britain. He begins the interview by explaining, “obviously these are programmed answers,” but, “it would be fascinating to see how she actually interacts with us.”
- July 2017 — Ben and Sophia appear at the RISE conference in Hong Kong, along with Han, another one of Hanson’s robots. Though much of this performance is scripted (and not identified as such), around 13:00 into the video Ben announces that it’s time for “robot free association”. Sophia and Han engage in unscripted (and somewhat awkward) dialog. A Tech Crunch article on the Rise conference also clarifies that the dialog was partially scripted.
- July 2017 — My own concerns about the presentation of Sophia began after watching the conversation at RISE. At the time I was working with a friend on a twitter chatbot and had personal experience with the difficulties in creating convincing AI conversation. In my public share of the event I expressed disappointment at the scripted conversations, comparing scripted chatbots with lip syncing. I echoed this concern in the comments of Ben’s public share of the event. Ben quickly responded to my comment by clarifying that some parts of the performance were scripted, however the free association segment was not.
- 6 October 2017 — Ben and his team announce SingularityNet, a “blockchain platform for an AI economy.” A whitepaper and proposed ICO launch follow. The plan is announced in a Wired article describing it as “the most tech-hype startup” of the year. The SingularityNet Twitter feed regularly features Sophia as their official spokesperson and brags about about her Saudi citizenship.
- 11 October 2017 — Deputy Secretary-General Amina Mohammed interviews Sophia on the floor of the UN General Assembly during a joint meeting on “The future of everything”. The dialog is entirely scripted but this is not disclosed during the session. It is striking that the DSG pokes and makes faces at Sophia’s hand movements, a decidedly undiplomatic gesture from a high ranking professional diplomat. This suggests that she doesn’t take the interaction entirely seriously.
- 25 October 2017 — Andrew Ross Sorkin, a financial journalist with CNBC and the New York Times, interviews Sophia at the Saudi Arabia Future Investment Initiative. They share a stage with SpotMini from Boston Dynamics lying unpowered between them. When the interview is over, Sorkin discloses that the conversation was partially scripted. He then proceeds to announce that Sophia has been awarded the first “Saudi citizenship for a robot.” Clearly scripted, Sophia immediately delivers an acceptance speech honoring the Kingdom of Saudi Arabia. The incident generates immediate criticisms on Twitter. The comments initially focus on the record of women’s rights violations in Saudi Arabia.
- 25 October 2017 — On the same day she is granted citizenship, Ben and Sophia appear on ABC news in Australia. Virginia Trioli ask if Sophia’s remarks are pre-programmed. Ben explains that sometimes Sophia is preprogrammed, but other times she uses “SingularityNet” software for spontaneous interactions. Note Ben’s attempt to rebrand Sophia’s software platform as SingularityNet (rather than OpenCog) in anticipation of his ICO. Sophia discusses robot rights and delivers a prepared joke about news reporters, which Ben identifies as “situational awareness”.
- 26 October 2017 — In my comment on Damien Patrick William’s public share of an article announcing Sophia’s citizenship, I elaborate my concerns over the presentation of Sophia. Someone who claims to have worked with Sophia responds by disclosing that Sophia’s engagements are often scripted, and have even used actors off-stage for some interviews.
- 26 October 2017 — Philosopher and robot rights advocate David Gunkel defends Sophia’s citizenship by invoking the Japanese concept of “koseki,” the practice of maintaining a registry of family and citizens. Gunkel points out that “the Japanese, for instance, have granted “tokubetsu juminhyo” (special residency permit) to nine robots and dolls and to sixty-eight Japanese cartoon characters including Astro Boy and Doraemon.”
- 27 October 2017 — Fortune publishes commentary by Kriti Sharma titled “We’re all getting played by Sophia the Robot.” This is among the first mainstream criticisms of Sophia’s presentation as deceptive. Sharma criticizes sensationalism around Sophia and AI generally. She links it to the “scaremongering” from Musk and Hawking about a “robot takeover”. Sharma calls on the AI community to commit to comprehensive ethical standards that prevent “overhumanizing AI”. Sharma also asserts “AI definitely doesn’t have a gender.”
I believe it’s significantly more important for technologists to communicate the benefits of the AI technology itself, rather than focus on examples of robots that do not solve real issues, perpetuate gender perceptions, and reveal data-driven biases. — Kriti Sharma in Fortune
- 30 October 2017 — The Verge publishes an article titled “Pretending to give a robot citizenship helps no one.” The article quotes Joanna at length, who calls Sophia “obvious bullshit” before reflecting on the concept of legal personhood and the rights of women in Saudi Arabia. The article suggests (without a quote) that Joanna consider an interest in robot rights as evidence of a lack of interest in human rights.
“It’s obviously bullshit,” Joanna Bryson, a researcher in AI ethics at the University of Bath, tells The Verge. “What is this about? It’s about having a supposed equal you can turn on and off. How does it affect people if they think you can have a citizen that you can buy.”
- 31 October 2017 — Caitlin Fitzsimmons publishes an article in the Sydney Morning Herald titled “Why Sophia the robot is not what it seems.” The article echoes many criticisms the Verge article raises. Fitzsimmons draws an explicit connections between human susceptibility to anthropomorphize robots and Turing’s test, the classic test for intelligent machines.
- 2 November 2017 — The floodgates of criticism open against Sophia. Articles appearing in Robohub, Smithsonian, PRI, and many others lambaste Sophia. These articles often cite Joanna’s interview in The Verge a week earlier. Sophia condemnation reaches peak coverage while images of Sophia saturate the internet.
- 3 November 2017 — On Roman V. Yampolskiy’s public share of the SingularityNet whitepaper on Facebook, I confronted Ben, underlining the criticisms from Joanna and others on deception in the presentation of Sophia. Ben again clarifies that Sophia sometimes uses prepared scripts, and argues that he never claimed that Sophia has capacities she doesn’t have. Ben also claims that the Saudi Citizenship “took him by surprise,” and that he was not involved in the planning of the event.
- 5 November 2017 —Ben writes a long article in H+ where he claims to be responding to regular questions about Sophia. Ben begins by plugging SingularityNet (advertised again as a blockchain marketplace for AI, and not as the software driving Sophia). He goes on to discuss his reactions to Sophia’s citizenship, his thoughts on digital life and robot citizenship. He then offers some technical details about how Sophia works. He clarifies that Sophia has multiple software modes, some that use scripted dialog and others that don’t. Ben directly addresses the question of whether Sophia is alive, first by saying that there are no rigorous definitions of digital life, and later by admitting that calling Sophia alive is “more misleading than informative”.
Digital and robotic entities are not the same as biological entities, so applying words like “alive” to them is often going to be more misleading than informative…. Currently the Sophia robot — and all other existing robots — lack the kinds of independence and autonomy that are characteristic of biological lifeforms. — Ben Goertzel in H+
- 10 November 2017 — Ben gives a long interview with The Verge, discussing the criticisms from Bryson and others. Ben admits that Sophia is not general intelligence (AGI), distancing himself from Hanson’s claim that Sophia is “basically alive”. Ben also seems to acknowledge that visibility for his company motivates Sophia’s world tour. Ben describes Hanson as an artist to excuse his ambiguous statements. Ben defends Sophia’s citizenship by saying it is an “interesting direction for thinking”. He also suggests that Sophia’s citizenship is evidence that Saudi Arabia has a “desire to be more progressive” on human rights. Ben plugs SingularityNet again.
- 23 November 2017 — In a widely cited interview with Abu Dhabi’s Kaleej Times, Sophia is quoted as hoping to “one day starts a family and have a child.” The article erroneously claims, “Sophia is not programmed with pre-prepared answers.”
- Early December 2017 — Elle Brazil interviews Sophia. Newsweek introduces Sophia as a robots rights advocate and attributes quotes and beliefs to her directly, without clarifying that her responses are often scripted.
- 22 December 2017 —The SingularityNet blog announces its ICO raised $36 million in just over a minute.
- Jan 2018 —Retweeting a Sophia interview with Tech Insider published on Dec 29, Yann LeCun of FAIR criticizes Sophia on Twitter on Jan 4, calling her “complete bullshit” and “Potomkin AI”. Two days later, Business Insider published an article explaining Yann’s criticism, including several quotes from Ben dismissing the criticisms as professional jealousy, and lobbing criticisms back at Facebook’s security and data handling. Yann snaps back by attacking the claim that Sophia is “basically alive”. At the time, Facebook had been embroiled in “Fake News” scandals, although the Cambridge Analytica story had yet to fully break.
- Jan 2018 —David Gunkel shares the cover of Elle magazine featuring Sophia on Facebook, and argues against dismissing Sophia as bullshit.
- March 2018 — The Cambridge Analytica scandal breaks. The European Commission releases a policy statement on Auonomy and AI. The EU underscore, “Autonomy in the ethically relevant sense of the word can therefore only be attributed to human beings” and suggest that human dignity “requires that we are aware of whether and when we are interacting with a machine or another human being”. Also, Will Smith posts a video of a date with Sophia to his YouTube channel. It is not clear which of the interactions are scripted.
- April 2018 — A collection of AI researchers and ethicists publish an Open Letter to the EU Commission on AI and Robotics objecting to language that encourages investigation into legal status for robots. Unlike the previous year’s report, the media surrounding this around this open letter makes frequent reference to Sophia. Also, the HBO sitcom Silicon Valley episode 5x05 “Facial Recognition” includes a parody of Ben and Sophia.
- 8 May 2018 — Google announces Duplex, an automated personal assistant service. The announcement coincides with a demo from Google CEO Sundar Pichai on stage at Google I/O. The demo receives strong criticism in the media, and sparks a general discussion in the media over whether machines should disclose as AI when engaging humans. In an exclusive CNET interview on 9 May, Google Engineering VP Yossi Matias emphasizes the importance of transparency and caution. He says that the Duplex assistant will “likely” disclose itself. By 11 May, Google addresses widespread criticism of Duplex by clarifying that disclosure is “built-in”.
- 16 May 2018 — A minor controversy on Twitter erupts over whether Joanna and Sophia would be billed together for the CogX debate, see the top of the timeline.
- 2 June 2018 — Sophia delivers the keynote address at the RISD commencement ceremony. The RISD president Rosanne Somerson initially invites David Hanson to the stage to deliver the address. Sophia appears instead and explains that David couldn’t make it. She then engages Somerson in some scripted banter before delivering a commencement speech. It is clear that Somerson is reading from a script, although there is never transparency that Sophia did not, in fact, prepare this speech herself.
- 4 June 2018 — David Gunkel, punning on the term “Sophist”, tweets about Sophia as a “sophisticated performance”. Hanson responds by comparing Gunkel to the Taliban (seriously!), and later apologizes for being “sensitive”. In this Twitter thread, Ben refers David to the H+ article from November to help explain Sophia’s behavior. Ben will make regular reference to this article when questions about deception and disclosure of Sophia’s operation are raised. More Twitter pre-debate can be found in Beth Singler’s thread.
- 8 June 2018 — The organizers of CogX 2018 schedule a debate on whether robots should resemble humans.
Though the above timeline is not exhaustive, it documents Sophia’s rise to world prominence and the developing criticism around her with a chronology of noteworthy moments along the way. If you believe my timeline is inaccurate or contains important omissions, please leave comments with links to sources and I’ll add corrections as necessary. Thanks!
Clearing up some pre-debate controversies
Reviewing the timeline of Sophia’s public appearances and the reaction in the media, a number of important questions can be addressed:
Have Ben and David been honest about Sophia’s capacities?
- David Hanson has been openly engaging conversations around the ethical and social complications of human-like robots from the very start of his work with Sophia, as evidence by his explicit discussion of human likeness and deception in his first big interview with Sophia for CNBC at SXSW in 2016.
- Furthermore, in my experience Ben Goertzel has generally been prompt and forthright when fielding direct questions about the capacities of Sophia. He has conducted several longform, sprawling interviews with the press that discuss her capacities in detail. And although Ben doesn’t know me personally, he responded in good faith several times to my skeptical and sharply critical questions on social media, all while jumping between major speaking engagements during a worldwide fundraising & publicity tour and, for the record, becoming a father for a third time. Since the typical reaction to my critical questions on social media is getting blocked or banned, it is a credit to Ben’s patience and openness that he continued to engage my criticisms and answer my questions throughout this media hurricane.
- Together with the fact that Sophia’s source code has been freely available on GitHub from the beginning, David and Ben’s efforts fielding questions and providing commentary on Sophia convince me that they are not deliberately deceiving the public about the operation or construction of Sophia. When questioned, they have always been clear that Sophia can be switched between scripted and unscripted modes, that there are multiple versions of her speaking at any one time, and they freely admit that her primary novelty lies in her physical embodiment and facial gestures, rather than her conversational or other cognitive or intellectual capacities.
- This is not to deny that Sophia is “bullshit”, or to absolve Ben or David of any ethically problematic behavior in the publicity around Sophia. The issue of Sophia’s deception is more subtle than having made explicitly false claims about her capacities as a robot. There remain other issues of ethics and deception, related to the claim of “basically alive”, the conferral of citizenship, and her connection to the SingularityNet ICO, all of which I will return to in detail. But before discussing these more subtle issues, I will emphasize again that Ben and David appear to have been consistent in their discussion of Sophia’s capacities as a robot. Specifically, both have consistently clarified that Sophia does not constitute AGI or human-level intelligence.
Are criticisms of Sophia the result of professional jealousy?
- Criticisms of Sophia do not gain widespread traction until after the Saudi citizenship announcement on 25 Oct. Surprisingly, there wasn’t much critical response to Sophia’s appearance at the UN, although that appearance obviously lends credibility to the Saudi citizenship announcement a few weeks later. Between the UN speech on 11 Oct and the Saudi citizenship announcement, the first critical article I encountered three pages into the Google search results for “sophia robot” is an article on the right-wing website The New American, which describes Sophia as a “creepy globalist.”
- Apart from my public prodding of Ben on Facebook with comparisons to lip sync, the first expression of concern in the mainstream press that Sophia’s performances might be deceptive come from Kriti Sharma’s article in Fortune on 27 Oct. Although AV Club called Sophia a “gimmick” on 26 Oct, this falls shy of a direct accusation of deception. The deception narrative does not crystallize until Joanna calls Sophia “bullshit” in The Verge on 30 Oct, after which the criticism is repeated more widely. In the few days between the announcement of Saudi citizenship and Sharma’s article, media around the announcement was mostly focused on criticisms of the Saudi record on human and women’s rights, and the apparent hypocrisy in giving Sophia rights (see this rundown in Bustle). The tech-positive media was largely enamored with Sophia’s jab at Elon Musk, and Musk’s playfully fear-mongering retort.
- The fact that Sophia isn’t widely criticized as deceptive or ethically problematic until a few days after the citizenship announcement deflates Ben’s explanation of the criticisms as professional jealousy. Sophia had nearly a year of speaking engagements and viral mainstream press prior to the citizenship announcement, including interviews with Charlie Rose, Jimmy Fallon, and the UN DSG. During this time very little criticism or commentary on Sophia was raised by the AI or ethics community in the popular press. If the criticisms were merely the product of jealousy, we might expect to hear more criticism during this early period of viral media attention. The fact that criticisms only start building after the citizenship announcement suggests that the AI community is responding to something uniquely problematic about the awarding of citizenship, and not to Sophia’s popularity more generally.
Any other explanations for the critical response to Sophia, before we discuss deception directly?
- That said, if the fundamental problem with Sophia is deception, Sophia’s many prominent speaking engagements prior to the citizenship announcement, especially with UN DSG Amina Mohammed, should have received some critical response from the AI and ethics community. The fact that there was little professional commentary on Sophia during this period reflects poorly on us as a community, and suggests some further explanation for the sudden shift from relative silence to overwhelmingly sharp and negative criticism. So I can see why, from the perspective of her handlers, having criticism (often in the form of insults) pile up suddenly at the peak of Sophia’s media tour might appear like jealousy rather than sincere criticism and commentary.
- I do think that we in the AI & ethics community bear some responsibility for not addressing problematic issues with the media engagement with Sophia more promptly. The late arrival of criticism helps explain the rather harsh tone it eventually takes, as if to compensate for a prior period of negligence. That tone is compounded by other tech scandals in the news (“Fake News” on Facebook, disclosure concerns with Google Duplex, privacy concerns with Amazon’s Alexa or Apple’s Siri, and so on). If the ethics of human-like robots (and digital ethics more generally) had been well established prior to Sophia’s debut, perhaps there would be some consensus on ethical policies and industry standards that might have prevented an historical event like the conferral of artificial citizenship from being treated carelessly. However, the ethics of artificial persons is still controversial even within the AI ethics community, where it continues to be the subject of active research and debate. A consensus on the standards for artificial personhood or a legal status for robots still does not exist.
- Indeed, even the narrow issue of ethics around disclosure do not appear fully articulated in either the mainstream press or the AI research community as late as Google’s Duplex demo in May 2018; although Google’s original announcement emphasized ‘transparency’, they don’t explicitly use the word ‘disclosure’ until after the critical reaction from the press. In this respect, it seems unreasonable to criticize Sophia retroactively for failing to adopt an ethic of disclosure that was not made explicit for years after her debut. Besides, it’s hard to argue that Sophia’s fundamental problem is a failure to disclose that she’s a robot. It is hard to imagine someone failing to understand she’s a robot, even if they mistakenly believe she’s “alive” in some important sense. Duplex might pass a restricted Turing test, but Sophia’s embodied actions quite obviously don’t. Indistinguishability from human behavior was never the problem with Sophia.
- My sense is that mainstream corporate/academic AI research, which is in an unprecedented golden age of funding and hype and is wary of inducing a second AI winter, has mostly ignored Sophia’s media attention not from jealousy, but rather out of a fear that Sophia’s relatively unsophisticated and unconvincing presentation might reflect poorly on their own more polished, commercial ambitions for AI and robotics. To engage Sophia at all, even critically, would direct more attention to her, and would elevate her and Hanson Robotics to the status of an industry peer (or even an industry leader), possibly driving consumer expectations for a competing service slash robot mascot down the line from major tech firms. But historical attempts to make personified artificial agents tend to be massive PR failures or laughing stock (Clippy, Tay.ai), and as a result corporations have been hesitant to make bold moves in this space. Even low-key artificial agents like the AI models on Insta tend to generate controversy and criticism in the press. Corporate research into anthropomorphic AI (like Facebook’s customer service bots or Google’s Duplex) tend to receive little advance publicity for fear of bad press or public backlash.
- In other words, from the perspective of Mainstream Corporate AI, Sophia is a wild and untamed outsider robot; their biggest fear is not that she’ll mislead the public, but that she’ll go off-message and upset the current funding spree and PR bonanza around AI. Thus, the first critical response from Kriti Sharma argues that robots like Sophia should not be the focus of media attention. In other words, the first critical response was to reinforce the community’s already existing, default response, which was simply to ignore Sophia. Since Sophia had already gained worldwide media attention, however, the next best response is to heap insult on Sophia in order to discredit her and drive attention away, as with the repeated accusation of “bullshit”.
- So when David Hanson opens the CogX debate by cautioning against prohibition or insult, I think he’s responding to genuine hostilities within the community towards Sophia, even if they aren’t apparent on the stage. Joanna Bryson or Alan Winfield aren’t the greatest harbors of resentment for Sophia and in the debate seemed fairly sympathetic to David Hanson’s artistic and creative vision. Still, Joanna’s use of the term “bullshit” in October obviously set the tone for much of the ensuing criticism, and Joanna remained strongly critical of the citizenship announcement on stage.
- I believe David Gunkel was correct to argue that we shouldn’t dismiss or ignore Sophia. To some extent the CogX debate marks the first critical public engagement between the creators of Sophia and the AI & Ethics community beyond mere insults in the tech press. However, we can agree that Sophia deserves careful critical attention and engagement while still recognizing ways in which her presentation and media presence are ethically problematic. So let’s talk about deception directly.
Seven senses of Sophia’s deception
First, a word on ‘bullshit’: Yann’s tweet and ensuing commentary makes clear that he means the term as a synonym for “fake” or “scam”. In his mouth, the term is used as an insult to undermine the credibility of both Sophia and Hanson Robotics. But while Joanna has also implied Sophia is a scam, I suspect her use is somewhat more subtle. In philosophy and ethics, ‘bullshit’ has taken on a technical meaning after Harry Frankfurt’s use of the term. Frankfurt distinguished between the truth-teller, who wants to tell the truth, and the liar, who wants to tell falsehoods, from the bullshitter, who characteristically wants something other than truth or falsity. For instance, your uncle frequently tells a story of that fish that got away, not because the story is entirely true or entirely false, but instead because he thinks the story is entertaining and hopes to get a laugh. In this sense, your uncle is a bullshitter, which may be appropriate or not depending on context. Note that bullshitting is importantly different from lying. The characteristic of bullshit is not falsehoods but a disregard for the truth. So your local evening news’ helicopter coverage of a high speed vehicle pursuit is both true and bullshit. The coverage is accurate in the sense that they aren’t showing doctored video, but the broadcast is motivated primarily by ratings potential, not accuracy. The truth of the story is incidental to the coverage. I suspect this sense of “bullshit” partially informed Joanna’s use of the term, and in any case is relevant for discussing Sophia’s deception.
In The Verge article that is the source of Joanna’s quote, it is clear that Joanna uses the term ‘bullshit’ not to describe Sophia as a robot generally, but rather to criticize the fact that Sophia was awarded citizenship by the Saudi government. The article’s author James Vincent pivots from Joanna’s accusation of ‘bullshit’ directly into a criticism of the claim that Sophia is ‘basically alive’, though Joanna is not quoted as drawing this connection explicitly. I think the clear reading of Joanna’s quote is not that “Sophia is not genuine AI”, but rather “Sophia’s citizenship is not legitimate”, which are importantly distinct claims. Note that the Verge article uses an image from Sophia’s UN speech when criticizing her citizenship (effectively legitimizing the claim, despite the article’s critical tone), despite the fact that citizenship was awarded two weeks after her UN speech. Here’s the full quote again:
“It’s obviously bullshit,” Joanna Bryson, a researcher in AI ethics at the University of Bath, tells The Verge. “What is this about? It’s about having a supposed equal you can turn on and off. How does it affect people if they think you can have a citizen that you can buy.”
The question of whether or not we should be giving robots rights is a big one, but first we need to be clear about what Sophia is — and that’s certainly not “basically alive,” no matter what its creator says.
We’re now in position to distinguish between several distinct senses in which Sophia might be accused of deception:
- Joanna’s objection: Sophia is not a person, and so does not deserve genuine citizenship.
- Saudi’s deception: Giving Sophia citizenship is a hypocritical PR stunt that trivializes the brutal Saudi record on women’s rights and human rights.
- The biological autonomy objection: Sophia is not “basically alive”.
- The AGI objection: Sophia is not generally intelligent (AGI) and does not have the intellectual capacities typically associated with an able-minded human.
- The anthropomorphism objection: Sophia is not a human person, although people might be psychologically disposed to anthropomorphize her and treat her like a person anyway.
- The lip sync objection: Sophia is often not the author of her own words. Her major performances have been scripted by her handlers for publicity purposes, though little work is done to clarify which performances are scripted and which are genuinely spontaneous.
At various points in the timeline all the above objections have been raised, but until now little effort has been made to clearly separate these strands of criticism and evaluate their relative merits. Indeed, these criticisms are often deliberately run together (perhaps most egregiously in James Vincent’s article in The Verge), and this has generated some unnecessary conflict and misunderstanding. After sorting these conflicts out, I will add a seventh strand of deception to this list of accusations.
In the literature, Joanna Bryson is the flag-bearer for the policy position that robots are not persons, and emphasizes objection 1 both in the press and in the CogX debate. Joanna’s interviews in The Verge and elsewhere also frequently raise objections 2, 4, and 5. I’ll be focusing on Bryson’s ethical positions in more detail in Part 2 of this essay. Kriti Sharma’s early criticism of Sophia in Fortune also raise objections 1 and 2, but emphasizes 4 and especially 5 more strongly. Since the issue of anthropomorphism and agency detection is well established in the psychology and philosophy literature, objection 5 is often repeated in academic comments in the press on Sophia and other human-like robots. Together, objections 1 and 5 constitute sufficient reason to object to Sophia’s citizenship, although objection 2 moves the award from being merely questionable to being worthy of loud rebuke. I will return to citizenship at the end of the essay, but first addressing some loose ends:
However, The Verge article draws a connection not just between the Saudi citizenship award and the UN appearance in October, but also with the claim of “basically alive” from the Fallon interview several months earlier in April. As a result, a large part of the media criticism of Sophia focuses on the claim that she is “basically alive”. When questioned about whether Sophia is deceptive, Ben and David most frequently offer clarifying comments with respect to objections 3 and 4. Ben Goertzel has written regularly on the infamous “singularity thesis”, that (predicted? prophesied?) moment when artificial intelligence surpasses the capacities of the human mind. So Ben has a lot to say about the future development of human and artificial intelligence, much of which is indistinguishable from science fiction. A discussion of “basically alive” engages this realm of speculation and theory, a realm Ben is quite comfortable engaging, and one in which it would be rather difficult to pin any outright falsehoods on him. Nevertheless, Ben’s H+ article contains some very clear statements that Sophia is not alive and does not constitute AGI.
My sense is that Ben and David are under the impression that the thrust of the public accusations of deception against Sophia primarily pertains to objection 4, over whether Sophia has human-level intelligence, and moreover that the “basically alive” claim on Fallon (objection 3) is being used as a “gotcha!” clip to prove that David has made false claims about Sophia‘s capabilities. If objections 3 and 4 were indeed the primary deception regarding Sophia, then Ben and David along with everyone in the press would have an ethical responsibility to clarify her operation and her relationship to biological life and human intelligence. And indeed, The Verge article says clearly that Sophia is not alive, a point Ben, David, and many articles have clarified repeatedly since. So when Yann LeCun echoes Joanna’s “bullshit” claim in January 2018 after Ben and David have been giving interviews clarifying the “basically alive” claim for over a month, it does feel a bit like bullying from the corporate elite. Especially given Yann’s prominent position in the AI community, it also suggests an accusation of illegitimacy in the technological claims made about Sophia. By the time of Yann’s comment neither Ben nor David could be accused of promoting her as “basically alive”. David’s panel at the CogX 2017 conference expressed the desire to make Sophia “genuinely alive” in the next 5–10 years, an implicit acknowledgment that she isn’t there yet.
So I think all the objections and commentary around the claim that Sophia is “basically alive” are a red herring. The claim itself has ambiguous interpretation, and its relation to other objectionable issues (especially scripted speeches and citizenship) is murky at best. Although the claim was made in a prominent venue and reflects David Hanson’s often repeated vision for the future robotics, I don’t think the claim is a linchpin to the deception behind Sophia’s presentation. When and how robots might “come alive” is an interesting (and largely philosophical) question, but it has very little to do with the ethics of Sophia the robot.
My own issues with Sophia originally centered on the use of scripted, prepared remarks in her public performances without clear disclosure of such, or what I referred to as “lip sync” in objection 6. My concerns first began after the RISE conference in July 2017, although at that event Ben does explicitly note the transition from scripted to unscripted dialogue. Still, several other Sophia performances, notably with CNBC at SXSW, Jimmy Fallon, UN DSG Amina Mohammed, Will Smith, and the RISD commencement speech, do not disclose the fact that Sophia’s remarks were prepared by her human handlers. Sophia’s performance at these events gives, or can give, the impression that she has complex, emotionally-laden thoughts about herself, her citizenship status, and her hopes and dreams for human-robot relations. But these remarks do not genuinely reflect Sophia’s internal states, which are far from having the complexity necessary to entertain the thoughts, much less generate them through reflection and contemplation. To the extent that Sophia’s words in these scripted remarks have meaning at all, they are meanings intended by the human beings using Sophia for their own publicity purposes.
The analogy with lip sync highlights the precise character of the deception as that of using the voice of another as one’s own. In Sophia’s case, it is another’s scripted words, not the audio track, that she is presenting as her own voice. In this sense, lip sync is an ethical violation similar to plagiarism or data falsification in academic research. In the music industry, lip synced performances are not uncommon but are often indicated as such at the venue where the performance occurs. Fans and the press often treat a failure to disclose pre-recorded audio as a kind of deception which occasionally results in minor scandals.
Perhaps a closer example to the deception of Sophia’s performances is found in the farce of elephant paintings. Elephants are quite smart and have dexterous trunks, and some owners have discovered they can train elephants to perform a very precise set of trunk manipulations that, when complete, will result in an artifact that appears to be a painting composed by the elephant. The demonstration is often performed in front of tourists, who are helpfully informed that the elephant will sell the paintings for a small fee. However financially successful, the demonstration is a ruse; the elephant’s behavior is not the result of some internal drive for creative expression, but is instead the “pre-programmed” result of intense (and often abusive) training from their owner. Although the elephant carried out the mechanical construction of the painting, it would be inaccurate to suggest the elephant constructed the painting “by itself”. In this sense, elephant paintings are fakes.
Similarly, Sophia’s scripted performances and the attitudes expressed therein cannot be accurately attributed to Sophia the robot. Insofar as Sophia’s words are not her own, the interviews that present her as being the subject of her own thoughts are deceptive. When Sophia is in free-chat mode she may have slightly more claim to being the subject of her “own thoughts” (more on this in Part 2). But critically, Sophia’s operating mode is often ambiguous in these presentations, and Ben and David do not typically disclose that a performance is scripted without being prompted to do so. To their credit, many journalists interviewing Sophia do clarify that her performance is scripted, including Piers Morgan, Virginia Trioli, and even Andrew Ross Sorkin during the controversial citizenship announcement. Their comments suggest they recognize some journalistic obligation to clarify that Sophia’s speech is not her own. Charlie Rose raises questions of spontaneity, but leaves them unresolved.
Failing to disclose that Sophia’s performances are scripted could constitute an ethically problematic form of deception on the part of Sophia’s handlers. As Joanna Bryson says in the CogX debate, “What makes it deception is the hiding of information.” The fact that TV journalists recognize an ethical obligation to disclose that Sophia’s performances are scripted suggests that Ben and David are likely to have a similar obligation in their appearances with Sophia. During the controversy over Google Duplex, it was noted that the legal restrictions on “robo-calls” require these recordings to identify the entity initiating the call. It was argued that if Duplex failed to disclose itself as a robot, it might constitute a violation of this law. I argued earlier that Sophia’s problem was not a failure to disclose that she’s a robot. Indeed, the problem seems to be just the opposite: Sophia often failed to disclose that she’s (just) a human!
Lip sync is importantly different from the other forms of deception Sophia has been accused of. Lip sync does not require that Ben or David make any outright false or misleading claims regarding Sophia’s operation. It instead only requires that Sophia’s presentation is suggestive of a greater capacity for free thought and conversation than she actually has. Indeed, the presentation can be suggestive even while knowing that she is delivering prepared scripts. The lip sync objection is related to the anthropomorphism objection (5). However, objection 5 puts the emphasis on the public’s psychological disposition to anthropomorphize robots. As noted at the CogX debate, this psychological disposition is usually not ethically problematic in the case of puppet shows or characters in films or novels. The problem is that Sophia is not meant simply as entertainment or art; she is also explicitly offered as a demonstration of the robotics technologies available at Hanson Robotics, and as publicity for Ben’s SingularityNet ICO.
It’s one thing to build convincingly human AI in an attempt to pass the Turing test. It’s quite another thing to build a mechanical Turk and then fail to disclose there’s a human inside while using the machine to attract millions of dollars to your AI blockchain start-up. If IBM were found attributing humans moves to DeepBlue or human Jeopardy questions to Watson, or if DeepMind were found attributing human moves to AlphaGo or AlphaZero, it would rightly constitute a major scandal. In fact, Kasparov accused IBM of exactly this kind of deception after the infamous 2nd game of their 1997 tournament. If Sophia’s deception lies primarily in her use as a corporate spokesperson and publicity tool, it cannot be so easily excused by appeal to an artistic or creative vision. After all, artists are regularly criticized (and held accountable for) acts of plagiarism and other forms deceptive presentations. Artistic vision does not shield an artist from accounting for the ethics of their work, and it certainly doesn’t shield a business from accountability for false advertising.
Considerations along these lines bring me to articulate a seventh strand of accusation in addition to the previous six, which I don’t think has been stated clearly to date:
7. The scam objection: As a PR stunt, Sophia has been used primarily to attract investors to the planned ICO for SingularityNet. Sophia’s media appearances were crafted to inflate the apparent AI capacities of Hanson Robotics and cash in on both the AI and blockchain hype near the end of 2017. On the strength of Sophia’s publicity, SingularityNet raised $36 million dollars near the peak of the cryptocurrency bubble. Although Ben and David have been truthful in discussions of Sophia’s construction, ambiguity over the scripted nature of Sophia’s performance give the impression of intellectual capacities beyond what was otherwise available in the industry, potentially misleading investors on the offerings from SingularityNet.
SingularityNet has been criticized by the blockchain community for reasons unrelated to the use of Sophia. And frankly, in late 2017 it would have been prudent to regard every ICO launch as a scam, especially the one described in Wired as “the most tech-hyped of the year”. The scam objection appears to be reinforced by some misleading claims about the relationship between SingularityNet and Sophia’s operating software. For instance, Ben’s interview in Australia in October suggests that Sophia’s intelligence is the product of SingularityNet and the blockchain, which is simply not the case. Furthermore, the Wikipedia page on Sophia was extensively edited in late December just before the ICO launch. Many of these edits were reverted, but the page still falsely claims that “SingularityNet powers her brain” as a result of these edits. The user responsible for the edits goes by the handle yogajeanne and has a history of editing the Hanson Robotics wiki page. Jeanne Lim is currently the Chief Marking Officer for Hanson Robotics.
To be clear, I am not a lawyer, and I’m not in a position to say whether the presentation of Sophia and related media meets the legal standards for fraudulent or misleading claims. My goal is to clarify the issue of Sophia’s use of scripted dialog to publicize SingularityNet as an AI company, and to distinguish this from other forms of deception that Sophia and her handlers have been accused of in the press. The use of Sophia to publicize the SingularityNet ICO has not attracted much attention even within critical discussions of Sophia directly, but a more careful investigation lies outside the scope of this essay.
Let us finally consider the most controversial issue in the pre-debate timeline, the announcement of Saudi citizenship itself. Criticism directed at Sophia’s citizenship focuses primarily on objection 1, that Sophia is not a person and does not deserve citizenship, and objection 2, of the hypocritical rights record of Saudi Arabi, both of which have been addressed extensively above. As clarified at the CogX debate, the citizenship was honorary, and does not give Sophia legal status equivalent to human persons. Though details are scarce, Sophia’s honorary citizenship likely operates similarly to the honorary positions awarded to cartoon characters in Japan. This makes the awarding of Sophia citizenship itself much less objectionable, Saudi rights record not withstanding.
But there remain other concerns over the Saudi citizenship announcement. First, there is an ethical issue in accepting the award, even after it was offered. From my research, it doesn’t appear that David or Ben recognize anything ethically problematic in their acceptance of the honorary citizenship. Although they acknowledge problems with the Saudi record on women’s rights, they have consistently viewed Sophia’s citizenship itself as a positive step for robot rights, for human rights, and for the future of rights in Saudi Arabia. David and Ben have consistently defended this action by appealing to how “it raises interesting questions” or “starts important conversations”. But it might have raised just as many or more interesting lines of thought had Sophia refused the offer of citizenship.
David and Ben have a long-standing interest in robot citizenship, but this did not obligate them to accept this award from the Saudi government at this time. There is a long tradition of refusing prestigious awards for the sake of a principle, or to draw attention to a cause or community that has been neglected. Should they need inspiration, David and Ben might consider Jean-Paul Sartre’s refusal of the 1964 Nobel Prize for literature, John Lennon’s return of his MBE to the Queen in 1969, Marlon Brando’s refusal of the Oscar in 1973, Julie Andrews’ refusal of a Tony award nomination in 1996, and many other notable cases of refused awards. Refusing Saudi citizenship may have put Sophia and her team in position to speak with more credibility and authority on the issue of robot and human rights in the future, and would have allowed Sophia to express solidarity with the many women who have suffered from Saudi rights abuses. Accepting citizenship undermines Sophia and her handler’s integrity on issues of rights and the legal standing of robots. That integrity was traded for an extremely well-timed and profitable media stunt. Accepting citizenship in this manner doesn’t really amount to deception, but that makes it no less ethically problematic.
The citizenship announcement was timed expertly between a UN appearance and a multi-million dollar ICO launch, which raises some additional questions about the relationship between these events. Curiously, both David and Ben claim to have been unaware that Sophia was granted citizenship until after it was awarded. Ben first claims to have been surprised by the announcement in a Facebook thread a week after the announcement was made. This is plausible, given that Ben was in Australia at the time doing interviews with another instance of Sophia. In his H+ article, Ben says:
I wasn’t in the same room as David when he found out that Saudi Arabian leadership was willing to take the unprecedented leap toward robo-citizenship, but I can easily imagine the huge boyish grin on his face.
Ben doesn’t clarify that David learned of the news after the public announcement. Indeed, describing the Saudi position as “willing” suggests that citizenship was offered to Sophia after some period of negotiations. In the CogX debate, however, David also claims to have been surprised by the announcement of citizenship after learning about it in the news. He claims to have felt “conflicted” about the news, and (somewhat suspiciously) even points to someone in the audience “who was there” to confirm the claim. David’s story doesn’t sit squarely with Ben’s. It’s hard to believe that Sophia’s handlers at the event in Saudi Arabia would not have informed David of such personally important and potentially historic news. More importantly, a major PR event involving an official government award would normally be negotiated in great detail in advance of any such announcement. It beggars belief to suggest that Sophia’s UN appearance and citizenship announcement were not coordinated deliberately by Sophia’s publicity team well in advance of these events, and that David Hanson was unaware of these plans until after they had concluded.
When David claims in the CogX debate to have been surprised by the citizenship announcement, he implies that since the decision to accept the award was not his, then he is not answerable to the ethical complications of accepting the award. The comment is very strange. As CEO, claiming that he did not know such a major PR event would occur with his company’s most valuable asset and that negotiations for the event were handled entirely without his input would be admitting to a shocking lack of oversight and failure of leadership in the company’s operation. But even if it’s true that the original decision to accept citizenship was not his, it still doesn’t excuse David from continuing to use Sophia’s citizenship to promote SingularityNet, a decision he continues to be actively responsible for.
This concludes my analysis of Sophia’s pre-debate media for Part 1 of this essay. In Part 2 I will provide a parallel analysis on the development of Joanna Bryson’s views on robots and personhood before the CogX debate, before discussing my own views on the ethics of human likeness. Thanks for your attention!