While a fully functioning artificial general intelligence might be a bit further away, an AI system that has the language learning and comprehension capabilities of humans will be ready within the next 10 years. In my opinion, this is an important hurdle in achieving artificial general intelligence because much of human behavior and knowledge is embedded in the way we communicate and interact with each other. We’ve already seen progress in this space with narrow AI systems being able to create blog posts or take exams or perform basic sentiment analysis. Companies like OpenAI have even built systems that can “create realistic images and art from a description in natural language,” which demonstrates a growing ability of AI to interpret language in a way that’s similar to humans – where we consume information that is assumed or contextual. In addition to improvements in AI algorithms, we’ll also see parallel improvements in computing power and growth in the amount of data/digital information that systems can utilize for training. Both factors will enable an AGI’s ability to understand natural language. We already live in a world where much of human communication occurs online and this trend will continue leading to exponentially more information for AI to learn from. The scale of information that will be available to train and develop AI on how humans communicate and interact will push the technology over the finish line – thus creating AI with human levels of language comprehension.
Given the above, I would argue that the probability of this occurring is close to certain – in which case the implications are broad. On the positive side, this will revolutionize human-computer interaction. If technology can engage with us socially and emotionally like humans, tools like AI assistants will be significantly more useful. In a practical sense, there will be no need for buttons, or remotes, or switches. Generally, anytime we interact with technology, there will be less friction in communicating what we want to happen and the technology executing that decision. In this future, you will never truly know if you’re talking to another human or a computer, and we will interact with machines just like we’re interacting with other humans (using voice or gestures, or expressions). Additionally, if we apply this AI within narrower scopes, it could automate interactions that we experience daily. Combined with other algorithms, an AI of this caliber could replace most instances we might need to communicate with another human. At a very basic level, it could take over customer service for example.
However, in theory, there’s no reason why it shouldn’t be able to teach and engage with students, or diagnose patients and provide bedside comfort, or create content and entertainment for humans to consume. Whether these implementations are positive or negative are to be seen as we’ll be cautious of incorporating AI in areas we deem emotional or those that have narrower moral boundaries.
While a complete natural language AI would improve on our lives in so many ways, there are equally as many opportunities for harm. We’re already seeing instances where, for example, a real speech by a world leader is almost indistinguishable from a fake, propaganda filled speech created by AI. If we extrapolate this out into the future, it will be possible to create entire identities for politicians, diplomats, leaders, and more with AI that can engage with other humans in multitude of formats. In essence, an entity could leverage a powerful AI such as this to replicate any individual and the manipulate it to suit their intentions or needs. However, at the same time, we would also be able to better detect bad actors online and in the real world – at scale – as AI systems would be able to understand the nuances between human intentions and whether they mean to harm others, etc. For example, currently, it’s incredibly difficult for an algorithm to define hate speech – there’s so much gray area that involves nuances in written language, cultural context, etc. However, I’d imagine an AI in the next ten years could solve for this with an error rate that’s close to zero.
These examples highlight only a few of the ways that a natural language AI will change our society in the next decade, but they provide a taste of how significant this innovation will be. While the negative implications of this technology may be daunting, we should still push for its development. AI is already being used (and will continue to be used) to create products, services, tools, etc. that we use every day. If we instead incorporate an AI with better natural language capabilities and emotional engagement into these systems, we will end up with better products and better experiences. In the case of preventing harm or negative impact, it seems that the biggest component we need to solve for is authenticity. Given today’s use cases for such an AI, being able to verify when an AI is an AI, and a human is a human should drastically improve our confidence in the technology as well as our ability to catch and reduce harm. I would imagine that this would still be a big component that would hold in the future. One of the difficulties with preventing the misuse of a powerful AI is that a lot of the core technology will most likely be open source. This can be valuable by facilitating a decentralization of power in some ways, but we might need to take the bold step of creating boundaries of use (either via regulation or technology) to prevent abuse. Because of how fast AI is progressing and how quickly we are producing digital information, we’re likely to see an AI that has mastered natural language very soon. This will shake up how we interact with machines but, more importantly, how we interact with each other and our environments
Meet Patel is from the Bay Area, California and spent the majority of his career before HBS working at Facebook where he focused on integrity topics like hate speech, bullying, election abuse, etc. He hopes to pursue entrepreneurship after school and is particularly interested in tough tech industries like space, health care, and energy.
0 Comments