Computer scientists have been working on making “smarter” computers for decades. The chess-playing feats of Deep Blue are long outdone, replaced by the likes of IBM’s WATSON, the world’s best Jeopardy player, and Deepmind’s deep Q-network, brain-inspired software that teaches itself to play computer games, so well in fact that it can even beat the best human players.
Computers keep out-performing us on more and more tasks but it never seems as if they are ‘like us’. There are still many tasks that humans can do easily that are difficult for computers (you may remember your last interaction with your mobile phone’s voice interaction system). But what aspects of these tasks determine what is easy and difficult for humans and computers? This division is increasingly becoming difficulty to see.
Computer pic via Shutterstock
No matter how good computers may become at specific tasks, such as playing chess, it doesn’t seem that they experience the world as we humans do. There is no subjective feeling to their being. They are not conscious. In comparison, we attribute consciousness even for the silliest YouTube-cats.
Recent advancements in the neuroscience of consciousness suggest that truly conscious machines can be feasibly constructed, and a new project will seriously test this possibility.
At the heart of these advancements is a new theory called Integrated Information Theory (IIT). IIT is unique in the unified way in which it accounts for many of the empirical facts about consciousness. For example, IIT explains why some parts of the brain can be removed without affecting consciousness while the removal of other parts completely abolishes consciousness. Using similar terms, IIT also explains why consciousness fades in dreamless sleep and under anesthesia.
But IIT goes further by providing an extensive mathematical account of consciousness, describing exactly what is required from a given system, whether biological, artificial or both, to be conscious.
The theory itself is detailed and requires some mathematical background to fully comprehend. Put simplistically, IIT claims that if a system can be in a large number of states (informational capacity), and these states are interwoven in some specific sense (integrated), then the system would have high integrated-information, termed Φ. IIT makes the bold claim that systems with high values of Φ will have conscious experience, and that as such subjective experience would exist whether or not it seems so from the outside.
According to IIT, modern digital computers do not possess much internal experience as they are not constructed in a way that allows for high values of Φ, while the complexity of human brains suitably achieves astronomically huge values of Φ. The difference in architecture explains why humans (and cats) are conscious but computers, no matter how intelligent, are not.
A research team in Japan takes this possibility seriously, and proposes to implement consciousness in artificial systems. The team is led by Dr Ryota Kanai at Araya Brain Imaging in Tokyo, joined by an Australian team led by Associate Professor Nao Tsuchiya, Monash University School of Psychological Sciences. This ambitious project is funded by Core Research for Evolutionary Science and Technology (CREST) program of Japan Science and Technology (JST), and will last five years with a value of approximately 3.4 million dollars.
The team is trying to achieve something beyond artificial intelligence: artificial “consciousness”. Such systems are expected to think and feel autonomously, as well as learn flexibly. While other researchers are trying to make intelligent systems, such as smart cars or houses with many interconnected sensors, the team’s goal is to construct systems with high values of integrated information, Φ. Such systems are predicted to act with the flexibility and spontaneity of conscious biological systems.
The project will initially focus on the consciousness of biological systems and evaluate how information is processed and integrated within these systems. The research will look at how the brain integrates information in various states of consciousness, such as during wakefulness, sleep and anaesthesia. Parallel research projects will construct neural networks that achieve high integrated information. The final stage of the research will seek to implement artificial consciousness in a robot. This robot will be tested in terms of its autonomous behaviour and flexibility.
Conscious machines will bring about a raft of challenges and applications that we are yet to even seriously explore. Research into this challenging field will inspire new design principles and learning mechanisms that may transform future technology. IIT may offer us a unique chance to explore a future with machines that not only think, but feel.