AI and Production: Overview
Over the last couple of years I’ve become more and more interested in the ways that the research being done into Artificial Intelligence (AI) might be applied to production and post production. In this article I’ll be giving an overview of what AI is at this stage of development, and what technologies are being used. Later articles will cover immediate and future applications and implications.
There are two schools of thought within the AI community: those who think (and are working toward) a fully autonomous, self-improving AI are distinct from those working toward making Artificial Intelligence a human assistant, romanticized in science fiction like the ship’s computer in Star Trek.
I admit, autonomous Artificial General Intelligence – that a ‘machine’ could successfully perform any intellectual task that a human being can – worries the heck out of me ever since I read Tim Urban’s two part article The AI Revolution: The Road to Superintelligence. That’s the worst-case scenario.
Right now there are a few AI applications that could really be truly considered artificial intelligence – AlphaGo, Sophia, self driving cars, and to a lesser extend personal assistants like Alexa, Siri, et al.
For all the advances of recent years, I don’t think we’re close to autonomous artificial general intelligence any time soon, but the human assistants are with us now, as I’ll be detailing in later articles.
Even without Artificial General Intelligence there are many advantages to its precursor: Learning Machines. Most of the applications we are seeing referred to as AI are really learning machines.
Obviously, if you can’t learn, you can’t think, so Learning Machines are a fundamental step towards autonomous AI, but they should not be considered to be Artificial Intelligence. According to Yves Bergquist post CES 2017, the distinction is:
An AI is a whole agent exhibiting an autonomous or semi-autonomous capability to reason about itself, its environment and its interests (usually this environment is narrow, such as training stocks or driving a car, which experts call “narrow AI”).
Even in this scenario, AI is still expected to be applied within narrow fields, like helping Philip find the best route to singing practice! At this stage of development, Siri (and other personal assistants) are more learning machine than full AI, which doesn’t stop them from being very useful.
Learning Machines are generally based on neural networks and learn by example, rather than by specific programming (such as we’ve used on our products along the way, like The Troubleshooter, or First Cuts). One type of learning machine examines a large number of examples and, based on the examples it sees and feedback on the results, generates its own way of achieving those results. Training continues until it is as accurate on new material, as it is on the training run(s).
This is the type of machine leaning has been used to replace white collar jobs in medical diagnosis of skin conditions, many legal applications from replacing paralegals to predicting court decisions, tax preparers, and playing poker among dozens of examples.
More advanced learning machines are ‘programmed’ with the desirable result, and they generate their own pattern recognition that gives the result they’ve been programmed to achieve. An example of this type of learning machine would be Jukedeck’s music composition learning machine that has taught itself to write music.
Jukedeck is a great example of how quickly these technologies are changing. Compare music composed by the October 2014 version which sounds like some rather bad 1980’s game music:
Skip forward two years and in August 2016, the machine is composing music good enough for background tracks or maybe a health or retirement commercial. Remember, this is a machine teaching itself how to compose music. These are great strides for just two year, and the rate of change is accelerating. Remember, this machine is teaching itself to compose music.
That same rate of change is what we hear in music composition is what will eventually happen in almost every other sphere, including post production organization shortly, and longer term, at least basic edits.
These are just examples of advanced learning machines for sure, but they are not AI. AI – according to Yves Bergquist – is much broader:
The practice of artificial intelligence also includes entire academic fields such as memory, knowledge representation, symbolic reasoning, machine heuristics, genetic programming, information theory, systems theory, decision theory and about a dozen other areas of research.
Even if we only focus on Learning Machines (a.k.a. machine learning), the benefits will be enormous, and mostly available now, or in the very near future. When we can also apply the autonomous reasoning capability of AI to the field of production, it’s hard to predict what the end result might be. We will almost certainly underestimate the changes it ill bring.
Steve Jobs famously referred to the Macintosh as a “bicycle for the mind” because, while a human wasn’t close to being the fastest animal, a human on a bicycle outpaces most animals. That’s the role I see for applied Artificial Intelligence (also known as weak or narrow AI): helpers that will be able to skyrocket our efficiency.
We don’t even need the AI to be applied to our field to benefit. For example, learning machines have been applied to fields like speech-to-text and image recognition, etc. We don’t need any complex programming skills because the result of someone else training the machine, and the machine learning, tools like speech-to-text and image recognition are available to all apps via a call to a remote computer.
The downside is that this will almost certainly cost jobs as corporations seek the most cost-effective way to manage their businesses. The combination of robotic production and applied AI controls will eliminate most manufacturing jobs. Semi-Autonomous robots are becoming more agile and are planned to be used for elderly care. Robots in all their forms – few are bipedal – benefit from applied AI to adapt to circumstances and locations around them.
But it’s not only s0-called ‘blue collar’ jobs that are being replaced. Most of the examples of machine learning outlined above are good middle-class jobs as paralegals, tax preparers or medical technicians. There is no doubt that the application of machine learning now, and autonomous artificial intelligence a little later, will cost jobs in production, particularly in post production.
In Part Two, I’ll be examining the tools we have at our disposal right now, and how they’re going to be used in the near future.
In Part Three, I’ll speculate on how we could use autonomous AI in post production, the implications of the inevitable loss of jobs that will entail, and the opportunities it will create.