A Friendly HAL: How AI May Make The Future of Work More Inclusive
by Saadia Muzaffar
“The future cannot be predicted, but futures can be invented. It was man’s ability to invent which has made human society what it is. The mental processes of inventions are still mysterious. They are rational but not logical, that is to say, not deductive.” ― Dennis Gabor, Inventing The Future
It sounds almost paradoxical to say that we cannot predict a future we are about to invent. Yet here we are, on the cusp of changes so profound they would have been considered unfathomable even twenty years ago. With the arrival of artificial intelligence and computers that we can speak not just to but with, those familiar old science fiction tropes have in the last few years finally started to become real. We can now talk to machines in almost the same way we would with one another. And though the effects of this technology will be undoubtedly wide-ranging, few fields will be as deeply and quickly affected as the workplace.
The advances of the last few years have been remarkable. Consider: When Alan Dix and Janet Finley published the third edition of Human-Computer Interaction in 2004, they noted that “perhaps the most attractive means of communicating with computers, at least at first glance, is by natural language.” Despite their excitement, though, Dix and Finley also had to offer the caveat that “unfortunately the ambiguity of natural language makes it very difficult for a machine to understand…. [and] it seems unlikely that a general natural language will be available for some time.” But fast forward less than 15 years and conversational interfaces are ubiquitous, popping up in places we wouldn’t have even imagined. With Siri and Alexa, we have real life versions of the Jetson’s robo-assistants, while Slack bots set up our meetings, alarms, and reminders. All told, it means that work is about to change significantly.
Yet amidst such rapid change, we are also grappling with a future that can sometimes feel as if it has arrived too soon, fomenting a mainstream discourse around AI “stealing jobs,” and images of a workplace that is cold, lifeless, or inhuman — or worse, less fair. But while it is true that a whole host of jobs will undergo massive transformation, just like they did in the industrial revolution, which will require committed public programme and policy revisions, such worries must be balanced by a perspective that does not keep us from imagining, and indeed inventing the future — one where the capabilities of these technologies are harnessed to make work not just more efficient, but more inclusive, empathetic, and creative.
Ada helps companies scale customer support with AI — see how here
If you’ve had the chance to interact with Apple’s Siri or Microsoft’s Cortana, you would be right to be excited about imagining what natural language and conversational interfaces mean for creating accessibility for all kinds of people. If getting an answer from technology does not rely on recollection of specific code-based commands to elicit a specific response, everyone can figure out their own way and pace of finding what they need.
Among the most exciting aspects of this ease of use stems from contextual analysis — that is, the capability of AI powered bots to personalize and recommend better results for searches, and to automate conversations that are repetitive. Because machine learning gets more and more focused the more often it is used, in time, the sheer number of searches will turn a well-built AI engine into a strong predictor of future trends. And while the speed, agility and precision will make inquiries about what to buy, where to travel, what services to select very efficient, it is also an area that will see a shift in the type of work people will end up performing alongside bots.
Benedict Evans, a mobile analyst and pundit at venture capital firm Andreessen Horowitz, defines the situation like this:
“Today you would spend hours or weeks in data analysis tools looking for the right criteria to find these, and you’d need people doing that work — sorting and resorting that Excel table and eyeballing for the weird result, metaphorically speaking, but with a million rows and a thousand columns. Machine learning offers the promise that a lot of very large and very boring analyses of data can be automated — not just running the search, but working out what the search should be to find the result you want.”
Put more plainly, it is not just that work will change, but how we define what counts as work worth doing. A good framework for understanding how AI will impact jobs is to thus look at where bots are inferior to humans and where they shine. For example: Bots are very good for doing one thing at a time when you know what you want. If you were to say to a conversational, AI powered bot, “help me plan a conference,” it currently wouldn’t be of a lot of help because there are far too many variables in what kind of conference you want and the many different ways you could deliver it. But if you were to ask a bot to recommend the best auditorium in a three square mile radius in the financial district of Adis Ababa that can host 500 attendees, the speed and accuracy and consistency of this bot in asking you the right questions and narrowing its search until it arrives at your most favoured solution will be light years ahead of any human.
This kind of capability is exciting on its own. Where it gets even more intriguing is that bots may also be designed to provide the same consistent and thorough service to people who are able-bodied as well as those who are, say, blind or deaf. Suddenly, scaling to a wide variety of skill levels, abilities, access and expertise becomes possible because machine learning’s capability to improve its interpretive accuracy the more its used means that young and old, variously-abled bodies, and new learners and experts, can potentially get their own ‘lane’ to operate in, so to speak. The same thing that might take one person many questions with to narrow down to a focused answer might be a two-question exchange for someone else — but a robust AI platform would guide both to a viable answer at their own pace, and without losing patience or affecting service quality. What’s more, in time, both of these outcomes would improve. Self-improvement is a huge plus in the world of AI-enabled tasks as the system would track the pattern of how quickly (or not!) it was able to help resolve the issues it was presented.
There is, however, a further upside to this potential for better accessibility: in making things easier for those with specific needs, AI chat can improve interactions for anyone. Jeremy Manna, former Community and AI Support expert at Public Mobile, counts himself a fan of chatbot driven customer service platforms for just that reason. Moreover, he points to a new category of jobs that will be critical in the bot-powered workplace: the people who will essentially manage a hybrid team of bots and people.
A bot manager’s role will be complex and detail oriented, and Manna says it will be heavily reliant on “data and analytics to understand what to look for and understanding key KPI’s to make sure the viability of the Chatbot for the business [stays positive].” Manna deployed Toronto-based Ada Support’s bot solution. He underscores the critical functions for people manning the bot manager role, and suggests there are a series of necessary questions that need to be asked: “How does the platform work? How does the bot work? How do we measure performance? What does ‘good’ look like?” he asks. The focus is, rightfully, always on the end-user: “What is really critical is how well customers react [to the interface] and what will the customer experience look like when engaging with the bot?” says Manna.
But in order to maintain that focus on how customers are being served, it’s important to have many voices at the table. When asked who he thought were the people missing in the discussion around AI, Manna emphatically responds: “non-technical folks who ultimately will be the main end-users of many of these AI products!”
One example of AI augmenting productivity and capability of non-technical users in a currently fairly technical realm would be RightClick.io, a bot driven “complete website builder” that creates a website from scratch. Hira Saeed, who took RightClick for a spin in her piece for ChatBot Magazine, added some human unpredictability to her interaction with the bot, to generate a surprisingly favourable reaction. “I tried to confuse the chatbot by asking weird things, which is my all time favorite thing to do with chatbots,” she said with a wink, “it didn’t break a sweat. No matter how many times I tried to change the topic, it dragged me back to the main thing [of building the website] intelligently.”
A lot of repetitive work that is currently performed by people can find a home in a bot-augmented workforce. Imagine other types of tasks like daily or weekly employee satisfaction surveys, logging tracking tickets for problem solving, follow-up on reminders for individuals’ assignments and reports, and an AI-powered version of the Microsoft paperclip that you can talk to and that would actually help resolve issues.
Automating drudgery or repetitive tasks takes on a different dimension, however, when those tasks present genuine risk. One of the most compelling uses of AI is to take over the kind of work that can reduce human exposure to dangerous or unpredictable yet unavoidable situations. Nidhi Chappell, director of Machine Learning at Intel, is a big believer in AI’s potential for not just replacing but amplifying human capabilities, especially in automating tedious or dangerous tasks. Properly designed AI products and services may mean the difference between life and death in dire situations that we are currently seeing around the world, from floods in Bangladesh and earthquakes in Mexico, to the hurricanes causing widespread damage in Puerto Rico. These are times when the human resources powering first responders and emergency services get stretched painfully thin, not only making access to these services difficult, but also straining the capabilities of those working long hours, struggling to do their jobs well under the duress of fatigue and barriers like language. In those moments, some automated, intelligent help may prove immensely useful.
This is so because AI is often capable of learning patterns and developing better, quicker, and smarter ways of resolving problems or performing a task through machine-learning. As such, one of its best use-cases is to step in where human intervention is limited because infrastructure has started to break down. Take Rescue.io., made by Square One Labs. It’s a bot that tackles the very real possibility that in many emergency situations, speaking on the phone is not a viable or safe option to report a crime or dangerous situation. Rescue defines itself as “911 for chat and SMS for college campuses.”
Justin Clegg, of Square One Labs says Rescue is “like a silent and pinpointed distress signal, via messaging.”
“During an emergency you can quickly alert your emergency contacts to what is happening and where you are so they can help,” he says. “There are many emergency situations where it isn’t safe to make a phone call. Rescue lets you tap a few buttons and alert your family and friends, without alerting others nearby you.”
The team that built the bot powered it via a constantly evolving and intelligent machine learning engine. Rescue can learn the “context of a situation, relay critical information and provide an accurate location for dispatchers and pre-selected emergency contacts”
It’s that sort of contextual response that occurs when applications are developed by people who have first-hand experiences in the problems these bots are trying to help solve. This idea of building a bot that is smart enough to assess influences in the environment that may be continuously in flux is a mission critical part of how AI powered solutions become more effective by learning from hundreds of real-life experiences. It’s a point AppLabb’s Joshi is quick to point out in our conversation about who needs to be a non-negotiable part of building an AI-augmented future. “More subject matter experts need to get involved along with data scientists,” he says. “AI will disrupt every industry, and experts will play a critical role.”
Those experts will have to do more than merely possess knowledge, though; they will also have to deploy it appropriately, and sensitively, too. As Megan Molteni notes in her Wired piece on the recent advent of chatbots as caretaking agents, “over the past few years, virtual help agents have taken on surprisingly sensitive jobs in modern society: Counseling Syrian refugees fleeing civil war; creating quiet spaces of contemplation for millions of Chinese living in densely populated cities; and helping Australians access national disability benefits.” AI does not simply have to be about the completion of mundane tasks; it can also be about responding to urgent human needs. There are few areas in which this is more clear than bot counselling. WoeBot, for example, is a talk-therapy bot created by a team of Stanford psychologists and AI experts. An unusual mixture of check-ins, inspiration, and empowering tracking-for-awareness pattern recognition, WoeBot aims to help people manage mental health on a daily basis, and for a price much cheaper than traditional in-person therapy. Given the high cost of access to professional therapists and the quickly disappearing viability of government support for these options for the vast majority of citizens, WoeBot can provide some level of assistance to many people struggling to manage common mental health challenges, and also help real therapists stay in touch with a larger roster of patients than they can fit into their limited days.
If artificial intelligence and machine learning may make future of work more inclusive, how might current companies begin to deploy such tech? Asked what advice they would give to businesses who want to get their first toe wet in the AI-powered landscape of today so they are ready and prepared for tomorrow, the experts from this piece had this to say.
Joshi suggests starting lean, i.e. to try and automate just the generic tasks to chatbot, such as hours of operations, FAQs etc. He says to tackling cold feet for most businesses should begin with a “chatbot to assist your customer service in becoming more efficient, instead of trying the replace your customer service entirely on day one.” Essentially: free your best workers to do more higher-level problem solving and creative work, and let the bots handle the tedious aspects of authentication, inquiry definition, system tests, and documentation.
Manna echoes similar advice by asking clients to consider their current in-house technical expertise level and choosing a solution type that is a good fit. “Who will be building/managing this product within the company? Does it require technical resources to build out functionality or is it simple enough to use for a non-technical support team?” he says.
In her seminal science fiction work Bloodchild and Other Stories, Octavia Butler says, “we’re all capable of climbing so much higher than we usually permit ourselves to suppose.”
Like any other technology before it, AI’s potential to help or hinder humanity is largely dependent on how deliberate the builders of this technology are towards patterns of bias, impact and fairness. The effervescent world of chatbots today gives us a window into the potential for good — that can help humans harness a future that may reduce our exposure to danger or tediousness, and unleash our creativity.
Visit Ada Support to see how companies are using AI today to automate customer support.
Originally published at ada.support.
Originally published at ada.support on November 9, 2017.