In the past 2–3 years Artificial Intelligence (AI) has gained immense popularity in the industry, media & academics. Every business wants to incorporate AI in its products, every new aspiring student wants to become an AI expert & the media can’t stop talking about it. Nations seem to be in an AI arms race & the nations which are feeling left behind in this field, are creating task forces to catch up-to speed. Tens of courses on AI are being offered, hundreds of books are published & countless articles are written (including this one). Not to forget the AI doomsday predictions. It feels like first, AI is going to take all of our jobs, then it’ll get super-intelligent & ultimately rule the world.
Take a step back, breathe.
It’s easy to get carried away & it’s much more tempting to follow the hype. However, not all hypes turn out to be successful. A good understanding of any tech, what value it provides, how it works and its pros-cons would give us enough knowledge to not only make the decision whether to ride the hype-wave, but also expand our knowledge-base to offer our opinions on the topic.
Let’s start with you. How would you define AI?
Which of the following systems, you think have AI:
- Chess playing computer.
- Language translation software (or robot).
- Self-driving car.
- Content Recommender
Before defining AI, let’s try to understand what is intelligence? I believe, you’ll agree with me that most of the humans are intelligent. What makes us intelligent? Is a cat intelligent? Are humans more intelligent than animals?
In my opinion, intelligence is mental property of a living thing using which it:
- Is aware of its goals.
- Can learn from its environment & increase its knowledge of the environment.
- Use this knowledge to come up with solutions to achieve its goals.
- Execute these solutions to achieve its goals.
By extension of the above discussion, AI is, intelligence, built by humans, incorporated into systems which are capable of coming up with their own solutions to achieve their objectives. That would be awesome! Do you know of any such system?
Let’s look at some other AI definitions.
There’s an old geeky joke on AI definition:
cool things that computers can’t do
If you take this definition seriously, then the realm of what computers can do is ever expanding and accelerating as we speak. So, the realm of AI would be ever expanding. For example, 70 years ago, we couldn’t play 3-D games on a computer, so 3-D games would fall under the realm of AI. Today, computers can play chess and Go against humans, which according to this definition, wouldn’t fall under the realm of AI (which, in my opinion, is a narrow AI system).
Based on Turing Test & Chinese Room Argument we can infer the following definition of AI:
machines imitating intelligent human behavior
In my opinion, this definition is relatively better however, this doesn’t capture the nature of Artificial General Intelligence (AGI). For example, today, it’s possible to build systems which can understand natural language and have a conversation with human beings, for example, Alexa. These systems are getting better and better as they interact with humans. However, this doesn’t mean that such systems have attained intelligence. For example, you can’t take the same system and make it to play chess or perform stock trading.
Elements of AI course gives the following definition:
autonomous and adaptive systems
In practical terms this is a better definition of AI, however, in my opinion it still doesn’t capture the essence of AI. Using machine learning, it is possible to create an autonomous system which performs better with time, for example, Netflix. Does it mean that such a system has attained intelligence?
Here’s my definition of AI, which helps me understand the field better:
A human artifact/system which has the ability to solve problems of a particular domain. Given a goal and success factors, the system is capable of breaking the problem in to smaller parts, create solutions, identify factors which optimize the success factors, identify risks and try its best to achieve the goal. Such a system should also know how to behave appropriately when it encounters an unknown situation. With appropriate training, such a system must be able to solve problems of a completely different domain. All the different scenarios and appropriate actions that system takes in a particular domain and environment, must not be programmed and such a system is able to infer appropriate actions based on exploring its environment.
With this definition, if ever such a system is built, then with appropriate training such a system can:
- Behave as a language translator.
- Can play a game of chess.
- Become your financial portfolio manager.
- Drive a car.
What do you think? Which of the AI definitions you like? Do you have your own definition of AI? Post in the comments!
In next post, we’ll cover the AI types & its sub-fields.
Cheers!
