Every respected company is using AI nowadays. Having artificial intelligence means you would get funded, smart people would work for you and you would be acquired by a cool tech giant in less than 24 months. Sounds great, doesn't it? Hey, hey, slow down a little!
Last Wednesday, September 28th, 2016 big news broke. Amazon, Google, Facebook, IBM and Microsoft have formed a 'partnership on AI' to increase public understanding on the topic and 'maximize societal benefits and tackle ethical concerns'.
Along with the coalition new companies are starting to emerge in the world of AI. Existing startups abandoned .com, .io, .ly and .biz and turned to .ai (make sure to claim your .ai domain today!). The word 'AI' is hot and if you have it mentioned on your Linkedin profile you sure are #1 target for recruiters.
Lets look at how AI works in very simple terms
According to the Oxford Dictionary definition, Artificial intelligence is 'the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages'. So, a solution that uses data to do something normally requiring human skill or supervision is considered an AI-powered solution. OK, and how does a machine learn how to simulate human action? There are 2 things generally required to train a machine:
- Training Data
- A Machine Learning Algorithm
Once you have built a machine learning algorithm based on your training data you can then develop an application that applies that same algorithm on new data and takes a human-like action based on this algorithm. Consider an example of Blue River Technology, an AI company that applies chemicals only on weeds and not on plants or soil. I don't know how exactly their algorithm works but I would imagine that they have lots of pictures of weed, plants and soil and they build a model that learns how to differentiate between three. So, they would first obtain classified pictures (perhaps hire someone who would label pictures as "weed", "plant" and "soil"), then train the model and build a software that "tells" a piece of hardware to release chemicals every time it recognizes a "weed" with a camera.
A human action here is the release of chemicals, which can be done by a machine but the decision to spray does no longer belong to a human, it is based on a scalable algorithm. So, this means that with development of AI we are taking people away from the decision making process, right?
How do you create a training set to train the model in the first place? Is the data with classified crops available online? Yes, but not in massive volume. One of the biggest challenges for companies is to collect accurately classified data for which they need experts in the domain. Perhaps, once you have training data you then don't really need human involvement anymore? Again, wrong. Who would audit your models and make sure they are up-to-date? People again. And, above all, of course you need a bunch of smart data scientists and an engineer to build models and software apps, but this is a different skillset.
So companies need investments to collect the data, hire people to build AI engine and a strong operations team who would manage potentially hundreds of different models and monitor their performance over time.
Why are so many companies now going through rebranding and a new positioning?
According to CBInsights Blog, AI related investments in 2014 and 2015 were 3 times higher than in 2011 - 2013 combined. Total investments in AI in 2015 reached $2.4B with the total number of 397 deals made. This is $6M per company in one year! Not hard to conclude that AI industry is financially attractive. Everyone wants to be funded, so companies are flooding this market.
Is AI a new concept?
No. In fact, Artificial Intelligence relies on many statistical concepts invented centuries ago. And wait, most of these concepts have been employed extensively in the academic community. So, arguably AI has been in existence for at least a few decades. In 1950 famous Alan Turing wrote a paper where he argued that if a machine could carry on a conversation and it would be indistinguishable from a conversation with a person, the machine would be considered "intelligent". In 1966 Joseph Weizenbaum wrote a program that passed Turing Test.
Some of the statistical concepts used to train machine-learning algorithms have been employed in academia even earlier than that. One of the most well known statistical methods we learned in college is Linear Regression that goes back to 19th century.
In the last 5 years a lot of progress has been made in applying AI concepts to business with companies like Google, IBM and Amazon leading the way. Press, Education, Recruitment and Investors became hyped up on the concept and being a PhD in Computer Vision somewhere at MIT is no longer what-is-it-again-you-are-doing cool but no-wonder-Google-is-fighting-for-you cool.
AI is still booming, but this ship is not for every sailor
I know it is very tempting to call yourself an AI business. You will go far with this statement but only for so long. As people get more familiar with what AI actually means and start learning what it is that your business is doing, they may quickly realize that your branding is nothing but smoke and mirrors. And what does this say about the CEO of the company that misuses AI just to get more attention? Either the CEO doesn't know anything about AI in which case his/her company won't get far anyway or, even worse, s/he knows what AI is and lies about using it.
AI is very cool
If you want to employ AI in your business I would only ask that you do it wisely. If you are a non-technical CEO, learn at least a few basic concepts, educate yourself on what technology leading AI companies use, what kind of algorithms they build and how it could be relevant to what you do. Never stop learning and you won't get replaced by a machine