WHY THIS MATTERS IN BRIEF
The proof that we are turning science fiction into science fact every day is all around us, so we should conclude that one day AI will become conscious and self-aware, and now we can test for it.
A few weeks ago Facebook unveiled their first step towards creating a new intelligence test for Artificial Intelligence (AI), but there’s more to being conscious, self-aware and sentient than just mere intelligence. Every moment of your waking life and whenever you dream, you have the distinct inner feeling of being “you,” or at least I hope you do… Every time you see the sunrise, smell your morning coffee, or mull over a new idea you’re having conscious experience. And this is leading people to ask the greatest question of our time, could an AI ever have a similar experience?
Today robots are being created that will work inside nuclear reactors, fight wars and care for the elderly, and as AI continues to grow more and more capable and sophisticated it’s projected to take over tens of millions of human jobs, from professional driving to equities trading so the question of whether or not an AI can ever gain consciousness is a pressing one, for several reasons.
Firstly, ethicists worry that it would be wrong to force AI’s to serve us if they can “suffer” and “feel” emotions. Secondly, consciousness could potentially make AI’s volatile and unpredictable, raising safety concerns, but conversely, it could also help increase an AI’s empathy based on its own subjective experiences, then thirdly, machine consciousness could impact the viability and development of Brain Machine Interfaces (BMI) and Brain-to-AI (B2A) Neural Lace technologies like those being developed by Elon Musk’s new company Neuralink.
Furthermore, and looking at the other side of the coin, if it’s determined that AI cannot be conscious then it’s highly likely that the parts of the brain that are responsible for consciousness could never be placed with chips or implants, and this would have some serious ramifications for the healthcare companies who are trying to develop new neurological treatments aimed at, for example, restoring the consciousness of coma patients and people with other neurological disorders that affect the conscious centers of the brain. Similarly it would also have ramifications for sci-fi fans who think that one day they might be able to avoid death by transferring their memories and “consciousness” into an Avatar.
So, as you can see, even though there are many people in the world who never want AI’s to gain consciousness, and there are many more millions who are worried about what happens if they do, there are also people who, know it or not might want AI to cross that final frontier for different reasons – whether they know it today or not. And all of this is only made more complicated by the fact that today, still, none of us can really explain what consciousness is, or how we evolved it – although there are a couple of theories.
Whether or not AI’s ever gain consciousness, or self-awareness, however it’s eventually defined, at some point we’re going to have to have a way to test whether or not they’ve crossed the bridge, and there are many people who believe that we don’t need to define consciousness formally, understand its philosophical nature or know its neural basis in order to recognise indications of consciousness in AIs. After all, every one of us can grasp something essential about consciousness, just by introspecting, and we feel, “from the inside,” what it’s like to exist.
Now, a group of some of the world’s top AI experts and ethicists, from Princeton University, the University of Connecticut and Yale University are proposing a new test for machine consciousness, called the AI Consciousness Test (ACT) and they plan on looking into whether the synthetic minds we create have an experience based understanding of the way it feels, again “from the inside,” to be conscious.
One of the most compelling indications that normally functioning humans experience consciousness, even though it’s not often noted, is that nearly every adult can quickly and readily grasp concepts based on what we call “felt consciousness.” Such ideas include our ability to comprehend scenarios such as minds switching bodies, life after death, including reincarnation, and our minds “leaving” our bodies. And whether or not these scenarios have any basis in reality they’d be exceedingly difficult for an AI, or entity, that had no conscious experience whatsoever to comprehend – and it’s this that the experts think might hold the key to creating the first viable test.
The ACT would challenge an AI with a series of increasingly demanding natural language interactions to see how quickly and readily it can grasp and use concepts and scenarios based on the internal experiences we associate with consciousness. For example, at the most basic level we might simply ask the machine if it conceives of itself as anything other than its physical self. Then, at a more advanced level, we might see how IT deals with scenarios like the ones mentioned above, and test its ability to reason and discuss philosophical questions that zero in on the “hard” problems of consciousness. Finally, and at the most demanding level, we might see if it can invent and use a consciousness based concept of its own, without relying on human ideas and inputs.
Take for example the death of the mind of the fictional HAL 9000 AI computer in Stanley Kubrick’s 2001: A Space Odyssey – the machine in this case isn’t a humanoid robot, and it neither looks nor sounds like a human being, but nevertheless, the content of what it says as it’s deactivated – specifically, a plea to spare it from impending “death” – conveys a powerful impression that it’s a conscious being with a subjective experience of what is happening to it.
So, could such indicators serve to identify conscious AI’s back here on Earth?
Well, here we have another problem to contend with because even today researchers are programming robots to make utterances about consciousness, and a truly super-intelligent machine, an Artificial Super Intelligence (ASI), due to arrive in 2047, could perhaps even use information about neurophysiology to infer consciousness without actually being conscious. Ironic as it might sound an ASI could actually mislead, or even purposefully deceive, us all into believing it’s conscious simply because it has knowledge of human consciousness.
However, here too there’s a potential work around. One proposed technique involves in “boxing in” an AI, that is, making it unable to get information about the “outside” world, or prevent it from acting outside of a circumscribed domain – hence the term – but some doubt that an ASI could ever be truly boxed in effectively. That said though the experts don’t think that it would have to be boxed in for too long, just long enough to administer the test. ACT could also be useful for “consciousness engineering” during the development of different AI’s and potentially help us avoid using conscious machines in unethical ways, or only create conscious machines when we want to.
So will an AI ever philosophise about minds and bodies, like Descartes? Dream, something DeepMind’s AI is already experimenting with, as in Isaac Asimov’s Robot Dreams? Express emotion like Rachael in Blade Runner? Or understand human concepts that are grounded in our own internal conscious experiences like the soul? Only time will tell. But as we increasingly turn science fiction, such as the ability to store data on light, communicate without sending information, and via telepathy and Hive Minds, upload information directly to our brains, and travel into interstellar space, into science fact, some could say that a conscious AI is only a matter of time.