For a better debate on Artificial Intelligence
You might have heard of Facebook shutting down chatbots because they decided to speak their own language. Or a creepy robot that looks like a human saying it will destroy humans. Such scary pieces of news that seemed to be something taken from a sci-fi move are actual news indeed. Hearing them when we have people like Elon Musk and Stephen Hawking talking about how Artificial Intelligence can represent a threat to humanity makes the debate about AI uses and its risks sometimes hysterical.
However, we must bear in mind alarming headlines spread much easier. The news mentioned above were exaggerated reports of the stories they covered. In the case of Sophia the robot, we can notice how her answers are somewhat scripted yet. In the case of Facebook chatbots, their test was successful actually. They just happened to communicate out of the grammatical standards of English language.
It is no surprise people fear AI. Thise news makes people fear something like HAL 9000 (2001: An Odyssey In Space) or Skynet (Terminator 3), threatening humankind comes up. But to have a more effective debate on Artificial Intelligence, its opportunities, and risks. We need to move beyond the panic of exaggerated news and have a clearer understanding of its current use cases and possibilities to the next years.
Artificial Intelligence Use Cases
Before anything else, when we talk about AI we must bear in mind this is a broad field with several different uses. Because of this, it’s hard to come with a clear and unquestionable definition of what is AI. As this text does not intend to be a technical one, I will show what my understanding of AI is through current use cases.
In the AI field, some of the most mentioned use cases come from Watson. The Artificial Intelligence from IBM, that gathers a huge amount of information and brings insights from it. Most news of Watson use cases come from Medicine, where it is helping doctors to diagnose their patients more accurately.
Much before AI became a buzzword like today, Facebook had it working on something simple and a bit creepy. Those suggestions of tagging yourself or a friend in a picture are nothing more than an AI application: image recognition. Algorithms detect a standard on the pixels that are defined as the face of a certain person and whenever this standard is detected, Facebook suggests you tag that certain person.
Still on Facebook, there is a growing field on chatbots: ‘robots’ which people can communicate in the typically written language, like this I am using to communicate with you. A field in which Nama is a leading company, with cases such as Poupinha — a chatbot on Facebook Messenger that assists people in regards to public services in the city of São Paulo. Which is an interesting case of people getting in touch with AI, even though sometimes they aren’t even aware they are not talking to a human being.
With the ever-growing volume of data we are generating, AI algorithms are being used to generate insights — this emerging field is known as Big Data. This leads to other applications, such as using algorithms to base the decision whether to grant loans or not to clients. Or more controversial uses, such as COMPAS — a software that evaluates the probability of a prisoner to relapse into crime. It is being used to support judges decision on whether to parole prisoners or not.
What comes next?
Imagining the future is a complicated and uncertain exercise. But it is also a necessary one. Either for us to foresight possibilities and also to foresight potential threats. Considering the current use cases mentioned above, there are some possibilities we can imagine for the next steps of AI.
In Medicine, where we already have Watson, we can imagine more progress coming. Thanks to the data that is collected from patients, being processed with AI, we can have from a better understanding on how patients interact with medicines to more personalized treatment.
Self-driving cars, which already apply AI with image recognition, have a room for improvement. We can have a deep transformation on how we enjoy the cities we live in thanks to that. Cars can get connected to each other and self-coordinate the traffic, for example. This could make traffic lights obsolete, with the traffic managed by the traffic itself.
If we cross the fields of Image Recognition and Big Data, we can think of improvements we are likely to see in preventive maintenance. Improved algorithms, working together with image recognition, can take maintenance to a new level and decrease the risks of technical stops.
With more developed conversational interfaces, we can imagine we are going beyond the current chatbots to something more. Better AI, with a better understanding of conversations can, give room for companion robots — something that will be very useful as the population of elder people grows.
There are several possibilities coming with the progressive adoption and development of AI. These possibilities are so exciting that it is hard to imagine AI won’t be ubiquitous within next years. But we cannot overlook the risks that come, and we must debate them so that we take the most of AI.
But there are challenges
Yuval Noah Harari, historian and best-seller writer, thinks we are heading to the end of humanism and towards dataism. That is, the belief that truth lies in data, which we are gathering and analyzing more and more. But if we are going to trust more in data and in what machines say, based on the data they have, we must be wary of the challenges.
When we talk about AI and automation, it is feared that there will be a massive job shortage. Actually, automation is a force that is reaching not only manual jobs but white collar jobs too. New jobs are coming up too, indeed. But they will need a different skill and mindset that the usuals we put in practice.
Our education system is a legacy of Industrial Revolution. But people capable of working inside an assembly line are unlikely to contribute in a world where AI will be ubiquitous. More important than memorizing the right answers will be our capacity of using our judgment based on the outputs AI will provide.
There is still another challenge. Perpahps a much more important and concrete one. And it is not intelligent robots that decide that may end up deciding to eliminate humankind. It is bias. In spite of being a critical threat, it is invisible and must be tackled. Even though we are talking about Artificial Intelligence, the applications are human-made. Thus, prone to biases that are unconsciously inputted by people who develop them.
Stories like the London hospital which used AI in the 1980’s to support their hiring processes are a good example of such biases. It was found out later the algorithm filtered out women and people with non-European surnames out of the processes. In the same way, COMPAS, mentioned earlier here, has received complaints of being biased against minorities. Both cases show how unfiltered trust in AI can reinforce our own biases.
It is hard to imagine these algorithms were deliberately biased to cause harm to certain people. It is more likely they absorbed such biases while being trained. And we must keep this in mind when discussing AI.
To reduce the risks of biased AI, human judgment will be important, like said above. More than simply relying on AI outputs, we must analyze, evaluate and audit them. Machines cannot decide for themselves and we must not blindly trust them. And the human judgment must be diverse too so that we can mitigate our own biases that are inputted in AI.
In this sense, we must be in touch with AI, instead of fear it due to the exaggerated news. And to get in touch with AI, I bring some suggestions below.
The first one is OpenAI — initiative led by Elon Musk, aiming safe development for General AI. Their website provides some papers on the findings of their research team. Also, if you look for something more hands-on, there is Azure — cloud-based application developed by Microsoft. In order to learn how to operate Azure, there are online free courses at Edx. And to close the suggestions here is a big list of AI applications: for personal use, for companies (here and here) and for specific industries.
There is a good story told in the book The Inevitable, by Kevin Kelly, that we can use to guide our relation to AI. The world watched closely the chess duel between Garry Kasparov and Deep Blue, a supercomputer developed by IBM (in which Watson is based). The world chess champion ended up defeated by the computer. But after the duel, when public interest dropped, Kasparov went deep into how to work together with the computer. As a result of that, a new modality of chess came up: the cyborg chess. Where humans play against each other, with computers suggesting moves that can be done. The players have the power to decide whether to follow the suggestion or to override it. Thanks to cyborg chess, the number of chess players increased significantly.
The rise of Artificial Intelligence is a big opportunity. We can waste it, legitimating our biases, but we can also use AI to put a spotlight on them and work to reduce the harm they cause. And we can get in touch with much other new things with the help of AI, just like cyborg chess. We have an opportunity which we must take to make a better tomorrow.
