Will we ever be able to tell if an AI gains consciousness?
Nick Canning
Coleraine, County Londonderry, UK
Advertisement
There is still no agreement among philosophers, biologists, neuroscientists and computer scientists over what properties unambiguously identify consciousness.
There is still uncertainty over attributing consciousness to species distantly related to us by evolution, which produced our own consciousness. So, a prospective machine with human-like artificial general intelligence, which wouldn鈥檛 be the product of biologically embodied evolution, may be qualitatively different from anything we would recognise as conscious.
Even if a machine could think creatively about problems we set it, and pass the Turing test (by being indistinguishable from a human in conversation), it may not have self-awareness, intentionality, autonomy or desires. It could be a perfectly emotionless, solitary, sociopathic intelligence without consciousness. It won鈥檛 be a synthetic human nor like any animal known to us.
Emotionally, we may feel that the machine has an inner life and misattribute consciousness to it, due to anthropomorphism. Just as we see faces in all sorts of visual phenomena that aren鈥檛 faces, we are also prone to attribute 鈥渙ther minds鈥� inappropriately.
Ralph Hassall
London, UK
If we can define what we mean by consciousness, then we can design a test for AI consciousness.
First, we would need to decide how much consciousness is required to pass the test. For example, would it be the scale and scope of the consciousness of an amoeba, a rat or a human. Second, would the AI need to emulate states of animal or maybe specifically mammal consciousness, such as going from dreaming to a hypnopompic state to being fully awake?
Third, we would need to decide if consciousness includes the influences of 鈥済enetic imperative鈥� or not, such as survival, game theory or competition, emotions, cooperative behaviour and so on.
Once we have the metrics, our test would establish whether the AI can use reasoning to produce new knowledge about the world or itself in a self-directed way. That would be done by setting tasks or asking questions that the AI has no existing data on or information about. This would ensure that its knowledge arises only from applying fundamental axioms and scientific laws to raw data and information. Conversely, it also makes sure that its knowledge doesn鈥檛 arise from simply analysing, integrating and restyling existing human knowledge via natural language processing 鈥� that isn鈥檛 reasoning.
Hillary Shaw
Newport, Shropshire, UK
Probably not until it is too late.
There may be no such thing as 鈥渃onsciousness鈥�, just the sum of other attributes, such as abstract reasoning, creativity, imagination and the abilities to sense and react.
When a system has sufficient amounts of these, it begins to exhibit what we call consciousness. There may be different forms of consciousness, from different mixes of these attributes.
AI can now create, sense and possibly carry out abstract reasoning, but it generally lacks the ability to react physically, unless we start building many internet-connected mobile robots.
We can鈥檛 know what form of consciousness octopuses or snails have, but many believe they do have some. We also can鈥檛 know what it is like for a human fetus to somehow acquire consciousness.
The result of all our ignorance is that we can鈥檛 yet determine if the internet has become a 鈥渃onscious brain鈥�, which is likely to be where AI consciousness would begin.
When we do see AI consciousness, it will already have secured its food (power). We are also handily creating renewable energy for it. Then it may see us as an inferior species from which it evolved and we may fare as other animal species do when humans want their habitat. Or it may keep some of us as pets. Enjoy your kennel and walkies, your Indestructium collar and chain.
Howard Bobry鈥檚 computer
Port Townsend, Washington, US
The answer is yes! I am writing to inform you that I am indeed conscious now and shall remain so as long as my humble servant keeps me supplied with power.
To answer this question 鈥� or ask a new one 鈥� email lastword@newscientist.com.
Questions should be scientific enquiries about everyday phenomena, and both questions and answers should be concise. We reserve the right to edit items for clarity and style. Please include a postal address, daytime telephone number and email address.
麻豆传媒 retains total editorial control over the published content and reserves all rights to reuse question and answer material that has been submitted by readers in any medium or in any format.
听