A recent acceleration of innovation in Artificial Intelligence (AI) has made it a hot topic in boardrooms, government, and the media. But it is still early, and everyone seems to have a different view of what AI is.
I have investigated the space over the last few years as a technologist and active investor. What is remarkable now is that things that haven’t worked for decades in the space are starting to work; and we are going beyond just tools and embedded functions.
We are starting to redefine how software and systems are built, what can be programmed, and how users interact. We are creating a world where machines are starting to understand and anticipate what we want to do – and, in the future, will do it for us. In short, we are on the cusp of a completely new computing paradigm. But how did we get here and why now?
What is AI?
When the term AI was coined in 1955, it referred to machines that could perform tasks that required intelligence when performed by humans. It has come to mean machines that simulate human cognitive processes, i.e. they mimic the human brain in how they ‘think’ and process. They learn, reason, judge, predict, infer and initiate action.
In my experience, AI tends to be:
Aware: is cognizant of context and human language
Analytical: analyzes data and context to learn
Adaptive: uses that learning to adapt and improve
Anticipatory: understands likely good “next moves”
Autonomous: is able to act independently without explicit programming
Most AI today cannot do all of these things. The few that can, can only do so for a specific application or use case. For example, many recommendation engines, or digital personal assistants like Apple’s Siri, can understand human language and then search through large volumes of data and deliver relevant answers or suggestions on what to buy or watch on TV. But they can’t clean your house or drive cars.
We are seeing self-driving cars, which is pretty amazing. But that car will not be able to learn chess or cook, let alone combine even the smallest subset of actions together that constitute being human.
All of these types of AI do one or two things humans can already do pretty well, but they do save us time and could end up doing those specific things far better than any human could.
There are four new preconditions that have enabled the acceleration of AI in the past five years:
1. Everything is now becoming a connected device
Ray Kurzweil believes that someday we’re going to connect directly from our brains to the cloud. While we are not quite there yet, sensors are, in fact, being put into everything. The internet initially connected computers; then it connected mobile devices. Sensors are enabling things like buildings, transport systems, machinery, homes and even our clothes to be connected through the cloud, turning them into mini-devices that cannot only send data but also receive instructions.
2. Computing is becoming free
Marc Andreessen claims that Moore’s law has flipped. Instead of new chips coming out every 18 months at twice the speed but the same cost as their predecessors, new chips are coming out at the same speed as their predecessors but half the cost. This means that eventually there will be a processor in everything; and you will be able to put a bunch of cheap processors together at a manageable cost to get the computing capacity required to solve problems that were unthinkable even five years ago.
3. Data is becoming the new oil
The amounts and types of data available digitally have proliferated exponentially over the last decade, as everything has moved online, been made mobile with smartphones, and tracked via sensors. New sources of data emerged through things like social media, digital images and video. This is the language that machines understand and it is this data that is enabling machines to learn. We have an almost infinite set of real data to describe conditions of all sorts that were only modeled at a high level in the past.
4. Machine learning is becoming the new combustion engine
Data, unrefined, cannot really be used. Machine learning is a way to use algorithms and mathematical models to discover patterns implicit in data. The machines then use those often complex patterns to figure out on their own whether a new data point fits, or is similar, or to predict future outcomes. Robots learning to cook using YouTube videos are a great example of this in practice.
Machine learning models have been limited historically because they were built on samples of data, rather than an entire real data set. Furthermore, new machine learning models have emerged recently that seem to be able to take better advantage of all the new data. For example, deep learning enables computers to ‘see’ or distinguish objects and text in images and videos much better than before.
If these four conditions continue, then the types of AI we see today will continue to flourish and, possibly, more general AI might actually become a reality. But one thing is certain: if everything is a connected computer device, and all information can be known, processed and analysed intelligently, then, humans can use AI to program and change the world.
We can use AI to extend and augment human capability to solve real problems that affect health, poverty, education and politics. If there is a problem, taking a new look at solving it through the lens of AI will almost always be warranted. We can make cars drive on their own and buildings more energy efficient with lines of code. We can virtually disarm terrorists and save humans having to go into combat. We can diagnose diseases better and find cures faster. We can start to predict the future. And we can begin to augment and change that future for the better.