Artificial general intelligence (AGI) is often hailed as a potential superintelligent technology that may match or exceed human cognitive abilities in every way. While the concept of such a technology is compelling, skeptics—both internal and external to the community creating these technologies—are cautious about setting expectations. In a survey of the Association for the Advancement of Artificial Intelligence, three-quarters of respondents said current methods of building AI technologies are not likely to be used to build AGI.
One limitation of AGI is the amount of material that exists for AI to train on. Given that AI mimics human thought and behaviour by interpreting patterns in human-generated information (for example, text, images, and sounds), its capacity to create original thought is bounded by the rate at which new information can be introduced and synthesized into the model.
It’s been reported that companies like OpenAI and Anthropic have incorporated almost all the English-language text that’s available on the Internet to their models. As a result, a reinforcement learning process is relied upon to strengthen the output of generative AI systems. This process involves learning through trial and error, using existing information to iteratively optimize performance. This approach has been applied in other use cases, such as in AlphaGo (a machine built by Google researchers to play a Chinese board game called Go).
The lack of consensus on how best to measure human intelligence is also a roadblock to benchmarking the ability of AGI to match human cognitive abilities. Because human intelligence can go beyond capturing and processing information from numbers and images, a true assessment of human abilities would need to capture aptitudes like emotional intelligence and intuition. In the workplace, socio-emotional skills are essential to performing many roles that involve personal interaction and require a greater amount of responsibility and decision-making, which sets a higher standard for what AI can reasonably be expected to do better than humans.
Because AGI is posited to have the ability to replace human capital, its ramifications for the labour market could include decreased wages, rising inequality, and unemployment. Mitigating the impacts of these trends would require restructuring our existing economic and societal systems—a transformation that institutions are currently underequipped to support.
Although many AI builders and enthusiasts see this technology as inevitable, there are mixed opinions about whether that point is near. But in light of recent layoff waves at companies that have started decreasing human capital in favour of AI, it is imperative, now more than ever, to future-proof the workforce.
Recognizing AGI’s potential impact and approach should provide an incentive for policy-makers and researchers to investigate solutions to alleviate the societal woes that may well come along with this technology.