REALIZATION OF A GENERAL-PURPOSE AI WITH THE FUNCTION OF CONSCIOUSNESS (1): OVERVIEW AND THREE HYPOTHESES
What kind of "consciousness" does a human being have - at ARAYA, we are continuing our research to answer this question. By creating an artificial equivalent of consciousness (AI), we are trying to get to the bottom of it and realize a "general-purpose AI".
In a four-part series, Ryota Kanai, President of the company, will explain how he plans to approach the project in concrete terms.
Realization of general-purpose AI with the function of "consciousness" (1): Overview and three hypotheses (this article)
Realization of general-purpose AI with the function of "consciousness" (2): Information generation theory - "imagining" an imaginary situation
Realization of general-purpose AI with the function of "consciousness" (3): Global workspace theory - Bridging information between modules of the brain
Realization of general-purpose AI with the function of consciousness (4): Meta-representation theory of qualia - Being able to cope with new tasks
HAVING A HUMAN 'CONSCIOUSNESS' WILL ENABLE 'GENERAL-PURPOSE AI' AND WILL GREATLY IMPROVE SOCIETY
We believe that when AI incorporates the function of consciousness, its capabilities will undergo a qualitative change.
Current ai is limited in what it can do. for example, it is useless if it goes beyond the scope of the knowledge it has learned. if you take a self-driving car that has been trained only on japanese roads and bring it to the us, it probably won't work. This is because traffic rules are very different. With current technology, Japanese self-driving ai will need to be re-trained with a lot of data tailored to the us environment before it can be used in the us.
In contrast, when a person who is used to driving in Japan tries to drive in the US, the car controls are basically the same, so as long as you pay attention to "stay in the right lane", you can drive quickly.
In fact, this is where the role of consciousness comes into play a little. We carry out many of our actions unconsciously. But this is because the environment around us is the same as it usually is. As soon as the environment changes, we become aware and act consciously. In other words, when we try to get around in a new situation, we are awakening our awareness.
After a thorough examination of the functioning of consciousness, we have come to the conclusion that the key to realising so-called general-purpose AI (general-purpose artificial intelligence) lies in the functioning of consciousness.
If we can embed the functions of consciousness into AI, we can expect to see self-driving AI that learnt how to drive in Japan adapt quickly to American traffic rules, as well as AI that can be used for a variety of other purposes, such as being an attentive housekeeping robot or providing business advice. If general-purpose AI becomes a reality, the benefits to society will be far greater than those of today's AI, and it could lead to innovation in all sorts of industries, from agriculture, forestry and fisheries to manufacturing, distribution, telecommunications, finance, medicine and education.
INCORPORATING CONSCIOUSNESS INTO AI ALLOWS IT TO "THINK CAREFULLY AND LOGICALLY" AND EVOLVE
We are not the only ones paying attention to the function of consciousness. Professor Joshua Benzio of the University of Montreal, Canada, one of the world's leading authorities on deep learning, has proposed the concept of "Consciousness Prior" (*1). This is an idea that would greatly enhance the capabilities of AI by incorporating a mechanism equivalent to consciousness.
1: Y. Bengio, "The Consciousness Prior," https://arxiv.org/abs/1709.08568
Professor Benzio likened today's AI to "System 1" (*2). System 1 refers to the reflexive thoughts that pop into our heads unconsciously. For example, just by hearing someone's voice on the phone, you can tell if they are angry. Today's AI, which excels at image and voice recognition, is only at this level, argues Professor Benzio.
2: Daniel Kahneman, translated by Akiko Murai, Fast and Slow, Hayakawa Shobo, 2012.
In contrast, the slow and logical mode of thinking is called 'system 2'. This mode of thinking is called 'system 2', and is used when planning complex tasks. There is no doubt that people are aware of what they are thinking when they do this, so if we want AI to evolve into system 2, we should add the capability of awareness.
THREE HYPOTHESES UNDERPINNING MAKING AI CONSCIOUS
How exactly can we give AI the function of consciousness? We believe that we should draw on theories from cognitive neuroscience about the mechanisms that give rise to consciousness. In fact, there is no standard theory yet that explains how human consciousness arises.
We are trying to apply the following three hypotheses to AI. We are trying to apply three hypotheses to AI: information generation theory, global workspace theory, and the meta-representation of qualia theory. Each of these hypotheses adds to the explanation of the function of consciousness, and could form the basis of a means of realising general purpose AI. Each of these will be discussed in the following articles.
Hypothesis of consciousness | Functions of consciousness | HOW TO ACHIEVE GENERAL PURPOSE AI |
---|---|---|
Information generation theory | Counterfactual virtual simulation | Flexible learning (Solution by Simulation) |
Global workspace theory | Connection of several functions via a common latent space | Solution by combination of task-specific AI |
Meta-representation theory of qualia | Objectification of functions using the space of meta-representations | Creation of a latent space representing the relationship between several neural nets (Solution by Generation) |
Click here to read more
Realization of a general-purpose AI with the function of consciousness (1): Overview and three hypotheses (this article)
Realization of general-purpose AI with the function of "consciousness" (2): Information generation theory - "imagining" an imaginary situation (next article)
Realization of general-purpose AI with the function of "consciousness" (3): Global workspace theory - Bridging information between brain modules
Realization of general-purpose AI with the function of consciousness (4): Meta-representation theory of qualia - Being able to cope with new tasks