Yesterday I attended an event titled ‘Are Machines Conscious? Debating AI’s Final Frontier’. I didn’t have to read much more of a description to sign up, I was intrigued. The event was a zoom call hosted by AI LA where a neuroscientist and a computer scientist debated whether machines could be conscious given what we know today. Those of us who signed up got to listen in, ask questions in the chat and take part in a vote based on the arguments presented. It was great to hear two points of view coming from different disciplines. It will come as no surprise that this topic has many unknowns and no definite answer but it did expose some potential paths to wander down. I also managed to learn a few new terms which I’ll share with you:
This is a term used in Philosophy to describe the qualities that are accessed with subjective experiences. If you look at an object, you may experience it differently to other objects and the things that contribute to that, are described as ‘qualia’. Some mental states are said to possess qualia like seeing, hearing, tasting, pain, emotions. The way things seem to us (our perception) when we experience those mental states is another definition. It might help to think of the contrasting experience, which is when you have a belief about something but don’t actually experience it. So in this debate it was brought up as part of the question, can machines have qualia? If a machine looks at a picture, it analyzes it, maybe it can pick out aspects we wouldn’t see as humans, but whether it has a subjective experience, I’m not sure.
Theory of Mind
This is the ability to attribute mental states — beliefs, intents, desires, emotions, knowledge, etc. — to oneself and to others.’ We develop a theory of mind at a young age when we discover from social interaction that other people have different feelings about things. It’s an important developmental milestone that helps us understand that while we think we know what another person is thinking, we only know it as a theory. We never know what another person’s mental state really is. So if we apply this to machines, can machines develop this same ‘theory of mind’. Can they detect that we each have different beliefs and intents well enough to know how to give the right response. We’re certainly trying. Think how powerful an AI would be if it could understand the subtle clues we humans recognize in each other. An AI with theory of mind. This article explores the idea of having a robot that possesses theory of mind and the new social capabilities that could be possible. However, one of the points brought up is that we are still struggling to understand exactly how theory of mind works in humans, so adding it to machines is a tall order.
Integrated Information Theory (IIT)
This is related to Qualia in that it deals with the ‘hard problem’ of subjective experiences. But it specifically deals with how a physical object generates a subjective experience in your mind. IIT claims that a level of consciousness can be measured with a property called ‘phi’. The ‘phi’ can be thought of as a type of synergy, which is calculated from the parts of a system and it adds up to something more than each part. This article is great as it gives an honest account of how confusing the theory can be. When applying it to real objects, it can lead to some odd conclusions that make little common sense. Nonetheless, it’s another subject that adds to the debate.
Is a Thermostat Conscious?
The discussion debated what consciousness is, mentioning the concepts above. But without a clear definition it’s easy to talk past each other and this happened quite a bit among the speakers. For example, is consciousness apparent in the things we are seeing in machines today like our AI assistants and chatbots or could a thermostat be capable of being conscious? After all it detects the temperature automatically and makes adjustments – it ‘knows’ something about its environment and reacts. A chatbot can illicit an emotional response from you. AI is hitting the heights of being able to pass the Turing test, and leave you confused to whether you are talking to machine or human. But something feels a bit unlikely about the claim that this amounts to consciousness. Thus we come back repeatedly to the question ‘what’s the definition of consciousness?’.
If a thermostat is not conscious, as we will probably conclude, what would it take to say a machine is? We are making good progress in AI so at some point, we could be dealing with a machine that is learning enough to break some kind of consciousness barrier. The challenge is figuring out where the threshold is. My understanding of the debate (and I can’t say I understood every part of it) was that we definitely haven’t agreed that threshold but it’s worth having the discussion as we start building better AI. The neuroscientist opinion was that we aren’t there yet but we can’t dismiss it’s possible. The computer science opinion was that we can believe we are there, if we widen the definition of consciousness to include the things AI does today.
The question I had but didn’t get a chance to ask, was ‘what are the implications, one way or the other?’. If we say, yes machines that learn and reach a certain level of intelligence can be branded conscious, we’ll need to make some decisions. Maybe one consequence is not being able to turn off those machines. Or making a bigger effort to instill the ethics and morals for machines. If we say no, are we in danger of waiting till it’s too late? In which case, an AI could reach a point of no return and make decisions driven by consciousness we hadn’t taken notice of.
It’s truly fascinating to hear the experts points of views and for that I’m grateful to have had the opportunity to attend. I hope we see a lot more of it. While we may not land an answer to the question, the continued debate could help us recognize the signs if we were to ever achieve such a level in machines. It also adds to the knowledge we are building in neuroscience about how our human brains work. Understanding what consciousness is, whether studying our own experience or designing a new one, is a good way to learn. Finally, it reiterates how important a knowledge of Philosophy is becoming in the development of Technology.