10 Things You Should Know If You Want to Build an AI
Playing with AI is like solving puzzles and playing board games these days. Thanks to the advancements, accessibility, and awareness, most people from engineering fields and other STEM fields would be tempted to try their hands at AI sometime in the times we are living in. You could be one of those curious people yourself, wondering what you should do if you want to build an AI yourself if you could create one yourself.
In my opinion, from my knowledge, these are what you should know before you build an artificial intelligence agent:
Goals: This is quite common with regard to any project. An aimless project is akin to opting for some random course in your college just because you think it might be “cool” or “easy” but without having any vision or interest or goal! That really doesn’t end well, does it? What do you want your AI to do? No, “everything” is not an answer. “If it really is intelligent, it should know what it is needed for right?” could be the next line, but well, … Sorry, you need to know what it is that your AI would do and what it is not intended to do. Is it over-simplistic, something that doesn’t even require “intelligence”, really? Or are you taking it to the other extreme, almost impossible to achieve in this lifetime?
Performance Measures: Okay, you have set a goal, so how would you rate whether your goals have been achieved or not? This part, evidently, varies from project to project. Imagine a class full of students, who are all taught a subject, and they are given exams. Now, how should a teacher rate their performance? How do you define the boundary that differentiates success from failure? If it’s not binary, then how do you define performance measures? How realistic is it?
Environment: It is not really smart to expect a hockey champion to win a swimming contest! What environment will your AI be working in? It is based on this factor and of course the goal that most of your training methods should be decided. This is also one major factor that is going to decide how complicated your artificial intelligence project is going to be.
Risk Awareness: Are you familiar with the popular Murphy’s law that states that if something can go wrong, it will go wrong? Nobody intends to make mistakes that would cost them much, but how much do you have covered? What are the possible flaws? How well have you studied the environment? What kind of, amount of knowledge do you have? What is the worst case scenario? Imagine excitedly toying with a big machine that can suddenly explode when it gets heated much. That’s like giving a driving license to a child or someone high on drugs! Are you equipped enough to handle your AI agent and is your AI agent trained well enough to deal carefully with all things that can go wrong?
Data: As I repeatedly keep saying, there is no artificial intelligence without data because there is no intelligence without knowledge. While knowledge and intelligence are two completely different things, intelligence requires knowledge in any case you can think of. And you equip your AI with relevant and enough data for it to be powerful and useful. Do you have enough data? Also, how reliable is your data? Using a salary dataset from the 80’s for prediction of salaries in 2020’s isn’t really going to help, right?
Algorithm: Which algorithms should you use and how would you customize them to train your model? This depends largely on your goals and the kind of data that you have. And that’s where the knowledge of algorithms would come in. How would decisions be made at each point? Which search algorithms would you use? Also, do you know how to pick the features?
Sensors: To get live data from its present environment, an artificial intelligence agent, such as a robot would need sensors. Take for example Ameca, a humanoid robot built for entertainment and exploration. The creator says that it has cameras in its eyes and that’s how it observes movement in front of it, so that the humanoid agent can even pull back if someone would come too close, getting into its “personal space”. He he!
Actuators: Not all AIs are virtual presence devices or just virtually conversational. We want AIs that can also do things for us, instead of just super fast calculating, predicting, and identifying. Take for example devices such as automatic vacuum cleaners or self-driving cars. These are automated devices powered by AI. And like all other mechanical devices that “do things”, for these automated, artificially intelligent devices, too, we need what are called actuators to help achieve the desired goal, such as cleaning or controlling movement. This humanoid supposedly has 27 + 5 motors in the face and neck!
Power: Well, yes, I did say “devices powered by AI”, but AI can’t literally offer power. How could your device be powered? What are the different options that are available and which ones can you afford? Batteries? Gas-power? Solar power? Depending on the kind of tasks you expect it to perform, the amount of power will vary. It is believed that for many self-driving cars, even GPUs fail because they have to process so much data! Now, are you looking at a pet project or a sample project or do you have a commercial, interesting idea in mind? For both the software and hardware needs, how much power would be required and what would be the best source for you? If it’s a small drone or a little robot that runs around the house, you could go with batteries easily available in the local market.
Programming: I was saving this for the last. This shouldn’t come as a surprise. Even the simplest applications require some coding knowledge at least. Arduino and Raspberry Pi are some of the most commonly used tools to help you do the coding for robots and drones. And these are really cool! How comfortable are you with programming?
“No one has a clue how to build a conscious machine, at all.”
— Stuart Russell
