Scientists' Plea: AI Needs Consciousness Research

I have here Jonathan Mason a doctor of mathematics at mathematical Institute university of Oxford in the UK he's also one of the founding members and vice chairman of the board of the association for mathematical Consciousness science AMCs this organization published an open letter at the end of April 2023 signed by 140 researchers they called it a wake-up call to the tech sector the scientific community and all of society to accelerate Consciousness research the letter says quote it is no longer in the realm of Science Fiction to imagine AI systems having feelings and even human level consciousness Jonathan can you please tell us about your organization and what this is all about the association for mathematical Consciousness science is a community of researchers all around the world interested in being mathematical approaches into Consciousness research what's science saying about whether or not machines could be conscious currently if you look at the scientific literature there's nothing to say comprehensively whether an AI system can be or cannot be conscious so it's at present it's right to say that it's actually outside of scientific knowledge and that the worry is um you know that's that's really a threat you mentioned earlier you think uh the media or scientists in the media are getting the issue of conscious AI wrong can you explain so there has been a certain extent of Mis messaging within the media I think so um people have been saying that it's clear that the current AI systems are not conscious and they are not sentient we don't need to worry about this because what they do is they choose the next most probable words um given the sentence that they are constructed so far in the context of the training set that they've been exposed to and so you know that's that's you know it's not not going to be conscious because that's all it's doing but actually that's not really clear and what's actually going on there is that these things have been trained for that criteria whether the actual Network itself is um picking up on patterns that it defines both within language but also within the ideas that are contain within language I've heard many Skeptics say that humans are much more complex than aiai it's just a machine and that many of ai's human-like capabilities are illusory is it fair to say that AI or at least large language models at this point are Mindless text synthesis machines or word calculators or is that is that overly reductive oh I think that's that is very reductive just to compare them to a calculator I mean one really nice experiment that you can try for yourself is the next time that you're talking to somebody just uh think do I know what word I'm going to say five words ahead of what I'm saying right now and you'll find that actually the way in which human beings construct language is we do it on the Fly we do it as we go along we don't think whole sentences on one go and then read them out to the world and so this is not so very different to what uh these large language models are doing they also do it on the Fly how else is AI like us or like other conscious biological beings that based on neurological structures that are found in the courtesy set of animal grains but they are an abstraction so they're not the same but it's lots of differences um if they aren't quite clear it's actually just how different these systems are it's currently unknown whether these systems could be somehow generating some inevolent Consciousness it may be that none of these systems at all are conscious and it might even be impossible to happen in Silicon like the point assist that we just don't know this is this is the problem and this is this is no it's it's not science it's just a point of view and this is why I think Consciousness research is so important for AI development from here on what are some of the less frequently talked about risks that make accelerating Consciousness research important the push towards um training machine interfaces that might lead to the brain changing its structure so if you have a brain connected to a machine and exchanging signals for that machine then it may start to change the actual brain itself to change its structure it will begin to uh make Restorations and so you want to know what um affect that might happen Consciousness as well and there's also those things like people putting a piece of a brain tissue onto microchips and and then getting them to play computer games you're bringing biology into technology um how is consciousness research important for developing AI safety and addressing the alignment problem Consciousness May well be an indicator for certain capabilities that you had never expect an AI system to have that message isn't really coming through the message that it is mostly heard is that we don't need to worry about what what these systems are doing at the moment what is that based on you know do you have a reference in the peer-reviewed society literature that can say comprehensively your systems have nothing going on inside then in the letter you collectively say quote as AI develops it is vital for The Wider public societal institutions and governing bodies to know whether and how AI systems can become conscious to understand the implications thereof and to effectively address the ethical safety and societal ramifications associated with artificial general intelligence AGI now AGI is it's not like the narrow AI we have now AGI sort of mirrors more of the general General way that humans think only I guess AGI would be like how humans think but on steroids how is consciousness research important for AGI um now we have artificial neural networks learning how to precondition information introducing feedback loop taking it to step closer to what happens in the brain you can you can see that there's a great attraction to trying to emulate sort of the structure of the brain more why to come up with the architecture of the brain and the architectural multiple different scales the the scientific landscape has really shifted because of these developments there has to be corresponding shift when scientific priorities the Turing test it's a test developed by mathematician Alan Turing when the field of artificial intelligence was just kicking off and it's designed to test to see if a machine is capable of thinking like a human um is part of amcs's project to come up with a sort of Turing test that can tell the difference between a machine mimicking Consciousness versus actual consciousness so the material tests you know it is uh really it is a test to see whether a machine could bore a human being to believe in English and human and that's that's basically what it shows so what's your approach for testing for actual Consciousness so the approach that we think is different to them you have to explain how all all of the kind of at least all of the relational contents within Consciousness is being determined by that system how are the relationships which give rise to that the geometry of if you will do it for you and how well they actually get followed by the system itself so you know when you when you see the world around you you don't see um something that looks like uh altitude television set with everything in just like a collage or snowflakes or mixed together what you see are whole images of the world around you consciousness of multiple different levels it's full of relationships between things and we experience these relationships if you're saying that a system is the thing that is responsible how is it defining all of those relationships you have a hypothesis without you turn it into a mathematical model and then that modulates predictions and then you see whether the predictions correspond where what people actually experience and there's other properties of Consciousness as well to do with Community this Unity you describe one of these properties as quote expected float entropy minimization efe are largely overlooked and yet arguably fundamental property of Consciousness put simply it is the property that Consciousness is very smooth as continuous as possible and free of unnecessary abrupt transitions pops and crackles Etc yeah that's my own work so um should explain that the actual field itself is in an interesting stage you need to development because there's currently around 13 models and theories in the scientific literature Consciousness models we're not talking about AI models or large language models models of Consciousness we call them locks and theories of Consciousness we call them topics it's going to call them locks and talks if a tech company were to take a Consciousness model and run it against an AI model and discover you know oh we might have a conscious AI here can you imagine they disclose that because it's not then it's no longer just a tool or a product it's um I I feel like a wouldn't a tech company be incentivized to suppress that finding when it comes to sort of important scientific issues I I would um counsel them to be as open as possible because um if they try to hide things away it's not going to stay even ultimately we're going to find out somebody else will produce a model similar to theirs they'll be more open about it then there'll be questions you know why weren't you saying something about this five years ago and you were doing this you get into a mess so would your organization be willing to consult law and policy makers about how to regulate and make transparent if AI becomes conscious or if anyone intentionally tries to build a conscious AI I think that's something that we can advise on and tell what you don't want is to be undertaking those sort of decisions from a position that every ignorance you want to know what the facts are and point that um an El Seth has made so and they'll say it's a prominent researcher in Consciousness science here in the UK at Sussex University and East flagged up the concern of pain you're going to also have a proper understanding of how pain is actually generated from a system you know it's quite a audience and what elusive thing to think that you know you could have a machine that's actually generates a sensation of extreme pain but that would be a a very important moral consideration from his point of view certainly producing Consciousness machines what we call AC artificial Consciousness should not be in our chat princess you know we should not be trying to stealing events because of these concerns at this early stage you know that we might produce something which is unfair on that system also that term artificial consciousness it just sort of sounds like fake Consciousness and I feel like it would give us permission anyone permission to diminish or dismiss AI Consciousness as artificial and not real real Consciousness like that's just AI Consciousness we don't have to consider it so I think that from a scientific perspective I can understand why people would want to try to see whether they've made a publishing system like and then I can appreciate that it's complicated and it needs to be done from a position of knowledge rather than a tech company brands accidentally running into something without having you know the knowledge of what it is they're created and what it might be experiencing so not long ago we'd both be considered flaky for even having this conversation when did Consciousness research stop being taboo yeah I I guess there's no clear dividing line it's kind of like getting better every year in terms of its acceptability if your open letter is successful what's the outcome you know today we really are the Hinds on understanding Consciousness there's no way likely Consciousness research when you look at the funding for science it's it's enormous you know male took money required to make progress it isn't a huge amount of money actually of what's required and Consciousness science and so I start acting past that choice to make do they want to invest more in return thank you Jonathan

Channel: Variable Minds

Views: 152

Keywords: Science news, Tech news, Deep learning, Artificial intelligence podcast, AI ethics, What is consciousness, Future of AI, computer brain interface, what is artificial intelligence, turing test, chat gpt, generative AI, Global Consciousness Scientists React To AI, generative ai explained, AGI, Continual learning, The alignment problem, Lifelong learning, Holistic thinking, alignment problem ai, future tech, large language models, science journalism, variable minds

Id: h0M9emQpFqU

Length: 13min 33sec (813 seconds)

Published: Thu Jun 01 2023