How far are we from sentient computers?

Artificial Intelligence (AI) has revolutionised all aspects of our life - from autonomous drones to self-driving cars and even personal assistants. Despite numerous impressive advancements, current AI is a pale imitation of human sentience, mainly due to its inability to understand human cognition or emotion. As Stuart Rusell, Professor of Computer Science at UC Berkeley explains, “a machine’s decisions may be ineffably stupid, in the eyes of an ordinary human, if it fails to understand human preferences”(Russell, 2021). 

For example, an AI system might be able to diagnose cancer in a patient, but it will not be able to understand the emotional toll it can take on the family. Similarly, a language model like ChatGPT would be able to provide objective answers to its user’s requests, but neither can it empathise with humans, nor understand the tone of human speech, like sarcasm or irony. Ultimately, the current limitations of AI in replicating human sentience highlight the need for further advancements in achieving human-like intelligence and consciousness.

To understand the extent of current AI's limitations, we must compare it with the complexity and depth of biological intelligence. However, devising a definition of intelligence for natural and artificial systems is difficult. Francois Chollet, a prominent AI researcher, has defined intelligence as "a measure of its skill-acquisition efficiency over a scope of tasks, concerning priors, experience, and generalisation difficulty" (Chollet, 2019). This definition captures an essential aspect of intelligence, the ability to learn new skills efficiently. Chollet's concept, meanwhile, falls short because consciousness encompasses more than just the learning of abilities. Moreover, emotional intelligence forms a vital part of interpreting emotions and responding in sensitive situations. Christie et al. (2015) in their paper talk about the controversy around including motivation as a prominent part of the emotional intelligence construct. Without a precise definition of intelligence, where all of its aspects are evaluated and considered, it is impossible to assess the current status of AI and pinpoint areas for development.

Sentience, a remarkable phenomenon, is the ability to experience emotions like pleasure or pain, interact with the world as an intelligent being, or think and acquire knowledge by itself is the frontier of true intelligence. Developing truly sentient AI presents a significant challenge due to the lack of consensus on the criteria to measure it. Although the intelligence of machines is often evaluated based on their ability to acquire new skills, even state-of-the-art AI systems struggle to acquire knowledge in novel situations. A self-driving car that is trained to navigate various road conditions and scenarios with high accuracy may struggle to deal with unexpected situations like a pedestrian suddenly crossing a street or detouring around a construction zone, and could not learn to read non-verbal cues from other drivers. Humans begin acquiring skills from the moment they are conscious, such as speaking and moving, which are essential aspects of consciousness itself. 

Another formidable challenge to achieving sentient AI is understanding the nature of consciousness itself and replicating human emotions. The nature of consciousness is subject to an ongoing debate, with some arguing that consciousness arises as physical processes in the brain and as a product of neural activity, or as a result of the environment as a whole. Growing evidence, however, supports that consciousness is “a product of the brain's ability to create a simplified model of the attentional state of the organism, including its sensory inputs and motor outputs” (Graziano, 2015). This suggests that emotions like pain and pleasure arise as a result of the brain’s ability to represent various sensory information. Take the example of how a dog learns to perform tricks in exchange for getting treats, or why humans naturally gravitate towards activities that give them happiness. Both of these tendencies have a neural connection to the result of “feeling” pleasure, which in humans, is marked by the release of dopamine. If machines are to exhibit this behaviour, they must experience emotions like pain and pleasure. We are still a long way from developing algorithms that can properly model and recreate the complexity of these emotions, as well as cognitive processes like perception and awareness. 

While current AI may fail to achieve human-like sentience, not all AI is created equal. Narrow AI is the name given to systems that “are unsuitable, for environments that offer little structure, consistency, rules or guidance” (Korteling  et al., 2021). It is the AI that we use in our day-to-day lives, whether it be virtual personal assistants or chess-playing AI. Narrow AI is good at a specific task, for which it is trained, and can perform at “superhuman levels” (Ibid.). On the other hand, the closest attainable approximation to sentient AI is a hypothetical form of AI, called Artificial General Intelligence (AGI). AGI aims to replicate the broad range of human intelligence and abilities and to even surpass human intelligence in every domain. Another work worth-mentioning in this context is that of Czerwinski (2021), where the team is trying to develop an “AI that feels” which can act as a virtual therapist for humans. However, the development of such technology is currently in its infancy, and it remains to be seen if AI can truly replicate human sentience. Albeit not anytime soon, development in this field of research could likely lead to sentient AI.

In conclusion, the development of sentient AI presents a significant challenge due to its current limitations, and the sheer complexity of human cognition and emotion. So far, Narrow AI has proven to be incredibly useful in certain domains, achieving true AI sentience requires a lot more research and experience, which we lack today. As we continue to explore the frontiers of AI and reimagine what it could hold for our future, the sentience of AI has the potential to be a revolutionary change in the AI field. Only by achieving true sentience can we create a future where AI not only aids humans but matches and even compliments or enhances human capabilities. The development of AGI could potentially lead to sentient AI, but it requires further research and exploration. Countless researchers, developers, and scholars have poured efforts into developing sentient AI, and we must bring their works to fruition.. Overall, creating machines that can empathise and comprehend human emotions is a daunting task, with an equally fruitful reward.