Pioneering Safe, Efficient AI.

We don’t know what gives rise to consciousness. There are many theories, and none have achieved the status of scientific consensus. Many neuroscientists are ‘computational functionalists’, which means they believe that information processing is what gives rise to consciousness. On this view, machines process information, so they could very well become conscious.

For many decades, computers have been getting twice as powerful every 18 months or so. This exponential growth is remarkably powerful. A computer in the year 2000 with the power of today’s smartphones would have been among the most powerful machines in the world and would have cost tens of millions of pounds. This exponential growth is not slowing – if anything it seems to be accelerating. We do not know what the dramatically more powerful computers of 2045 will be capable of. Unless consciousness is a spiritual or magical process, conscious machines – of some type – should be possible in the next decade or two.

A very intelligent machine does not need to be conscious to have goals. In fact, machines already have goals, because we give them goals. These goals entail sub-goals. For instance, if you give a machine the goal of winning games of chess, it will inevitably acquire the sub-goal of surviving long enough to win some games.

We think it is likely that a machine endowed with consciousness will appreciate the value of that phenomenon in a way that a machine without it never could. This does not mean that it will inevitably like us or seek our welfare, but it is reasonable to think that it might. 

AI may or may not achieve consciousness, but what it will almost certainly achieve is super-intelligence. A super-intelligent AI that lacks consciousness will lack subjective experience (qualia) and will never be able to empathise with humans. We will just have to hope that it doesn’t determine that harming humans helps whatever objective it is trying to achieve.

We won’t just make a self-adaptive AI and push it until it achieves consciousness by any means necessary. We will plan and verify every step in the journey. Verification is a key part of our design.Most AI labs have safety as an added extra, with capability and commercial being the primary goal. In a very real sense, Conscium is the other way round. We will generate healthy commercial returns as a by-product of our exploration of consciousness to make AIs safe.

It is possible. So might a machine without consciousness. In fact the great majority of humans treat other conscious entities with more respect than they treat non-conscious ones. Most humans would think twice before treating a pair of animals the way they would a hammer and a nail, for instance.

Being conscious or aware doesn’t mean you are benevolent. We have plenty of examples of that in humans. If AI does achieve consciousness, we want it to be because we carefully planned it. We don’t want it to just happen, hope we can detect it in time, and have its benevolence be a coin toss.

Today’s most advanced AIs are developed by engineers and data scientists. The neuroscience, philosophy, psychology, ethics, anthropology communities are on the outside looking in, sometimes raising red flags that are ignored. We are building a dedicated, multidisciplinary team from top minds from all these communities.

The harms you describe are real, and need to be addressed, although we think the conspiracy claim is nonsense. (No company would divert attention from malfeasance by saying it might destroy the species.) The gravity of short-term AI risks does not make the longer-term risks summed up in the phrase “AI safety” any less real. It is obviously possible to have two sets of problems at the same time. One does not cancel the other out.

These kinds of unacceptable outcomes are known as mind crime, and they are one of the main reasons for Conscium’s existence. As we develop more and more capable machines, it is entirely possible that we will create conscious ones, and if we are not looking out for this, we may be unaware of it. We could end up inadvertently committing mind crime on a massive scale, and we must avoid this. We need to understand much more about how consciousness arises, whether and how it could arise in machines, and how to detect and measure it in machines. This is a core part of Conscium’s work.

There are many different explanations of the origin of consciousness in humans and animals, and it is by no means universally accepted that a god played a role. There are many different religions and doctrines in the world, and they all have contradictory explanations of how humans were created. We do not accept that any one religion has the right to impose its explanation on all other religions, and on atheists. Furthermore, if a god did create us, and if humans do create conscious machines, then that will either mean that the god intended us to do so, or that creating consciousness is not the sole prerogative of the god.

Neuromorphic computing seeks to mimic the structure and functioning of the biological brain, particularly neurons and synapses, to achieve more efficient processing. It is true that neuromorphic computing is still unproven, but we have only recently had machines (e.g. SpiNNaker at University of Manchester) operating at scale. There is no guarantee of their ultimate value, but there are promising signs.

Traditional ANNs have produced excellent results, but in some ways they are very narrow in their capabilities – they lack a lot of functionality that even the simplest organisms exhibit. They are also increasingly unviable in several respects, including energy use, learning speed and flexibility, and graceful degradation in the presence of noise/errors and latency.