Will You Survive the AI Apocalypse?

“Just to rub it in, a version of AlphaGO, called AlphaZero recently learned to trounce AlphaGo at Go, and also to trounce Stockfish (the world’s best chess program, far better than any human) and Elmo (the world’s best shongi program, also better than any human). AlphaZero did all this in one day.” - Stuart Russell, Human Compatible

I was reading “Human Compatible” this week and the above anecdote got me thinking. A computer crushing Chess and Go Grandmasters is impressive and feels ominous, but what does it mean for our everyday jobs? Every year computer chips get smaller and faster (Moore’s Law) and experts predict Machine Learning, AI and automation will eviscerate our jobs. Is this the harbinger of the AI apocalypse or our sci-fi soaked imagination running wild? The answer lies somewhere in the middle.

Here’s a quick primer - the AI revolution has two distinct components: hardware and software. The first programmable computer, the Ferranti Mark 1 could perform about 1000 operations per second. Fast forward to 2019, the Summit supercomputer performs a thousand trillion operations per second, approximately ten times faster than a meagre human brain. But here’s the rub: faster isn’t necessarily better, all it means is you’ll arrive at a wrong answer quicker. Also, interesting sidebar: the Summit uses about a million times more power than a human brain. While we have the hardware chops (and I’ve not even mentioned quantum computing), do we have the requisite software programming skills to extract the best performance out of these behemoths? The reason AI such as AlphaZero (or AlphaGo and IBM’s Watson before that) perform these incredible feats of “intelligence” is due to extremely optimized, problem-specific programming that took the brightest computer-science brains years to develop. But for solving tasks other than Chess or Go, it’s no better. This is the paradigm of problem-specific vs general-purpose AI - it’s when leaps in general-purpose AI occur that we should be getting fidgety. To use a metaphor, a horse can outrun us in almost all scenarios (problem-specific) but is terrible at writing screenplays or making music (general-purpose). Leaps in problem-specific AI makes headline-grabbing news; leaps in general-purpose AI herald tangible, economic change.

I’ve devised a simple 10 question test for you to check if your job is replaceable by AI, based on the limitations of problem-specific AI and the characteristics of tasks that general-purpose AI can solve. Assign a score between 0 and 5 for each of the questions, tally them up and look at the results at the end of the quiz. Good luck! P.S., I am no AI-expert. Heck, I’ve only read a book about it and came up with this hare-brained test. So, the results are a bit of fun, not Gospel - take with a pound of salt.

For the questions below assume the task in hand is something you spend a significant percentage of time of your working day on - the results apply to this task and not your entire job.

1. Do you have all the specifics (information, data, tools, resources, etc.) needed to complete the task? Score 0 if you never do, 5 if you always do.
2. While completing a task, what percentage of the time are you actively working on it? Score 0 if you dedicate 100% of your time to completing the task and 5 if you never do i.e. you have a supervisory role, or it runs in the background.
3. Whilst completing a task, what percentage of time do you to interact with other people? Score 0 if you interact or discuss with other people 100% of the time, 5 if you never have to.
4. Whilst completing a task, what percentage of time do you use non-computing machines (for example tools, vehicles, equipment, supplies etc)? Score 0 if you always use them and 5 if you never do.
5. Whilst completing a task, what percentage of time do you interact with other humans via email, text, chat or other remote ways to complete the task? Score 0 if always and 5 if never.
6. Are the outcomes of your task predictable or not? Score 0 if completely unpredictable, 5 if completely predictable.
7. Does the nature of your task change or remain static over time? Score 0 if changing, 5 if static.
8. How much time do you have to make decisions in your task? Score 0 if you have less than an hour and 5 if you have a few days or more.
9. How quickly will you know if something has gone wrong with the task you completed? Score 0 if you’ll know in a few moments and 5 if you’ll know within a few days up to a year.
10. How process-driven is your task? Score 0 if not process-driven at all and 5 if completely process-driven.

Results

• A score of exactly 0: Congrats - barring death or retirement, you’ll keep your job forever.
• A score 25: Less than a 50% chance of AI replacing you within your lifetime.
• A score between 26 and 35: A 50 to 70% chance of AI replacing you within your lifetime.
• A score >35: A higher than 70% chance of AI replacing you within your lifetime.
• A score of exactly 50: Hello AI - thanks for completing this test.

My own score was 35 and as a result, you can hit me up on LinkedIn. Respond with your own scores below as it will make for interesting reading!

So, there you have it. There’s no doubt the AI apocalypse will come, and some jobs are more at risk than others, but the key takeaway is that progress in problem-specific AI makes headline-grabbing news while progress in general-purpose AI brings about economic change. Progress in the latter will take many more years to come to fruition and in the meantime, be aided by leaps in the former. Advances in general-purpose AI will bring about radically new workforces, novel new job-types, more automated tasks and more human-centric decision making. Yet, we tend to always assume the worst when projecting into the future. So, until Skynet embarks on a hiring spree, keep calm and carry on.