Hundreds of thousands of developers and data scientists are now working on AI. The pace of development is accelerating, with increasingly eye catching breakthroughs being announced on a daily basis. This has led to a lot of confusion about what we can do with AI today and what might be possible at some point in the future.
Here we have seven possible stages through which AI might develop, highlighting the types of applications we might see in the next 15-20 years:
Seven stages in the future evolution of Artificial Intelligence
To provide some clarity on the possible development path, we see seven distinct stages in the evolution of AI’s capabilities:
Stage 1 – Rule Based Systems – These now surround us in everything from business software (RPA) and domestic appliances through to aircraft autopilots. They are the most common manifestations of AI in the world today.
Stage 2 – Context Awareness and Retention – These algorithms build up a body of information about the specific domain they are being applied in. They are trained on the knowledge and experience of the best humans, and their knowledge base can be updated as new situations and queries arise. The most common manifestations include chatbots- often used in frontline customer enquiry handling - and the “roboadvisors” that are helping with everything from suggesting the right oil for your motorbike through to investment advice.
Stage 3 – Domain Specific Expertise – These systems can develop expertise in a specific domain that extends beyond the capability of humans because of the sheer volume of information they can access to make each decision. We have seen their use in applications such as cancer diagnosis. Perhaps the most commonly sighted example is Google Deepmind’s AlphaGo. The system was given a set of learning rules and objective of winning and then it taught itself how to play Go with human support to nudge it back on course when it made poor decisions. Go reportedly has more moves than there are atoms in the universe – so you cannot teach it in the same way as you might with a chess playing program. In March 2016, AlphaGo defeated the 18-time world Go champion Lee Sedol by four games to one.
The following year, AlphaGo Zero was created, and given no guidance or human support. Equipped only with her learning rules, she watched thousands of Go games and developed her own strategies. After three days she took on AlphaGo and won by 100 games to nil. Such applications are an example of what is possible when machines can acquire human scale intelligence. However, at present they are limited to one domain and currently, AlphaGo Zero would forget what she knows about playing Go if you started to teach her how to spot fraudulent transactions in an accounting audit.
Stage 4 – Reasoning Machines – These algorithms have a “theory of mind” - some ability to attribute mental states to themselves and others e.g. they have a sense of beliefs, intentions, knowledge, and how their own logic works. Hence, they have the capacity to reason, negotiate, and interact with humans and other machines. Such algorithms are currently at the development stage, but we can expect to see them in commercial applications in the next few years.
Stage 5 – Self Aware Systems / Artificial General Intelligence (AGI) - This is the goal of many working in the AI field – creating systems with human like intelligence. No such applications are in evidence today, however some say we could see them in as little as five years, while others believe we may never truly achieve this level of machine intelligence. There are many examples of AGI in the popular media ranging from HAL the ship computer in 2001 A Space Odyssey, to the “Synths” in the television series Humans. For decades now, writers and directors have tried to convey a world where the machines can function at a similar level to humans.
Stage 6 – Artificial SuperIntelligence (ASI) – This is the notion of developing AI algorithms that are capable of outperforming the smartest of humans in every domain. Clearly, it is hard to articulate what the capabilities might be of something that exceeds human intelligence, but we could imagine ASI solving current world problems such as hunger and dangerous climate change. Such systems might also invent new fields of science, redesign economic systems, and evolve wholly new models of governance. Again, expert views vary as to when and whether such a capability might ever be possible, but few think we will see it in the next decade. Films like Her and Ex Machina providing interesting depictions of the possibilities in a world where our technology might outsmart us.
Stage 7 – Singularity and Transcendence – This is the notion that the exponential development path enabled by ASI could lead to a massive expansion in human capability. We might one day be sufficiently augmented and enhanced such that humans could connect our brains to each other and to a future successor of the current internet. This “hive mind” would allow us to share ideas, solve problems collectively, and even give others access to our dreams as observers or participants. Taking things a stage further, we might also transcend the limits of the human body and connect to other forms of intelligence on the planet – animals, plants, weather systems, and the natural environment. Some proponents of the singularity such as Ray Kurzweil, Google’s Director of Engineering, suggest that we could see the Singularity happen by 2045 as a result of exponential rates of progress across a range of science and technology disciplines. Others argue fervently that it is simply impossible and that we will never be able to capture and digitise human consciousness.
Envisioning a Smarter World
To help bring to life what a smarter future might look like, below we have outlined a possible development timeline for AI over the next 20 years or so:
Now to 2020
Conclusion - Preparing for A Smarter World
The pace of AI development seems likely to continue and we can expect to see regular breakthroughs that blow our collective minds. However, we shouldn’t assume a smooth progression through the seven stages of AI outlined above. Moving from reasoning systems to AGI isn’t equivalent to the exponential progression we have become accustomed to in computer power, memory storage, and internet connection speeds. To reach our final three stages of AI, massive breakthroughs are required in areas such as neuroscience, understanding consciousness, neural networks, and deep learning algorithms.
What is clear is that, over the next 15-20 years, our world is likely to experience several fundamental transformations as this “Fourth Industrial Revolution” powered by smart machines touches every, nation, life, and sector on the planet. The critical priority and challenge is ensuring that these advances don’t progress unchecked and render humans irrelevant. We have to establish governance frameworks and a public voice that are strong enough to guide the development of AI in service of humanity not in place of it.
By Rohit Talwar, Steve Wells, Alexandra Whittington, April Koury, and Helen Calle, of Fast Future