I have written about AI before, and I am fascinated by the potential it brings for us to enhance our capabilities. But with every technological advancement, there are unintended side-effects that need to be recognised and addressed - and addressed they will be, either by design, or via conflict, if our history is a good measure for predicting the future.
The convenience of patterns
In 2003 I signed up for a service called AudioScrobbler. I had all my CD's digitised as MP3's (later, when storage became cheaper, I redid the process and stored my CD's as FLAC - but that is another story), and AudioScrobbler would keep track of what tracks I listened to, gave me a list of favourite artists, and suggested "Musical Neighbours" - people with a similar taste. Before long, the algorithm started suggesting artists, and I could even have recommended radio, at which point AudioScrobbler rebranded itself as Last.FM. I discovered more music that I liked, based on my listening profile (disclosure - I no longer have my Last.FM account).
Amazon is famous for its algorithms suggesting your next purchase - people that purchased this book also bought (fill in your book title here). And often, I found myself buying that second book as well. Never regretted it.
Since Spotify launched in Australia, I have been a very active subscriber (I found the 320 Kbit/s undetectably different in sound and experience compared to my 1000+ Kbit/s FLAC files), and I started a library with all my artists, and found that it grew fast (with millions of songs in the library, there is more to choose from than I would ever be able to listen to). Spotify has a Discover playlist - a lists with songs I might like based on my listening habits. Recently, Spotify introduced daily playlists - six playlists representing the different moods, with songs from my library. Convenience! But it also made me a lazy in the discovery of music. I explored more within a certain sound or mood, and less outside of it.
The irrelevance of fact
News would be consumed via TV, Radio, or Newspapers. The programs you watched or listened to, the papers you read, these were an indication of your convictions in the political spectrum. These mass media outlets needed to present the facts, and the interpretation of these to provide direction. The facts remained the same, and verifiable, the interpretation is what divided one news outlet from another. But it was still tailored to a mass - not an individual. Combine the large audience for consumption with the relatively small pool of outlets that had the resources to communicate, the internal competitive dynamics kept the media outlets in check. The size of each newspaper was tailored to the content a person could consume during a fixed portion of the day, usually morning over a coffee. The number of facts presented were small enough for them to be verified.
Enter the internet: content creation and dissemination are commoditised. Everyone can create content about anything and make it accessible to everyone else at very low cost. Verifying each story is impossible, or at best, utterly futile - in a stack of news, fake or real, who can prove the verification is true? It is also irrelevant.
Content is what glues us to the screen and keyboard. It drives us to interact, whether it is on LinkedIn, Twitter, YouTube, or Facebook. The more we interact on their platform, the better it is for these businesses. Applying similar algorithms as described above in the examples of Spotify, Amazon, and Last.FM to content, and you can glue the audience to your platform. Facebook suggests news, Google suggests news, YouTube provides the next video, and creates a selection for you. The AI behind this has one job - maximise the screen time per user. It will learn to provide you with the articles and videos you will consume, and it will learn to do this better and better over time.
The profit of your outrage
Facts are totally indifferent to your or my feelings, that’s why they are facts, not opinions. The interpretation of these facts may soften the blow or reinforce our position. An AI does not care about facts. It does not know how to tell fact from fiction (unless we tell it how to verify, and that may not be that easy). An AI does not care about your feelings either - after all, it is just a machine. It has no concept of the complexities that make us human. It does have an outcome it needs to achieve - and it will learn to maximise this outcome. If the outcome is screen time on a platform, and the variable is the content it feeds you to achieve this, then that is exactly what it will do: it will feed you whatever keeps you on the platform the longest, and if you could share and interact, all the better. It turns out, news that outrages us is very effective when it comes to interaction and sharing. The more outraged we are, the more we comment, share and display our discontent. That in turn is shared, and there will be some that are outraged by our outrage, and so on, and so forth. All the while the AI is seeing the improvement in the attainment of the goal that has been set, and the profits are coming in. With some poetic license, mainstream commercial AI has become the monster that feeds on our outrage.
The current tangible effects on society, and what to do next?
The middle ground, mainstream consensus is not profitable when it comes to engagement on platforms. As such, we have become more and more divisive in our communications. The formation of echo-chambers in the form of followers on twitter and the like (an interesting term: follower... if you follow each other, then there is no leader, rather likeminded people). News is curated to individuals based on what is most likely to make you interact - it will present news on certain topics you are interested in, and likely provoke a reaction. That constant reinforcement is providing a deep division.
What is also happening is that the traditional news sources are seeing a vocal divide, and some have jumped onto a side on issues to align with a potential customer base. What is interesting is to find out if that is done by means of personal outrage (conviction) or on commercial grounds (it probably is a bit of both). The fact is, our society is more polarised at the edges, and these edges seem to grow bigger at the expense of the "silent majority". And we luckily it is still a majority, because if this amount of outrage was mainstream, we would be in the middle of a civil war. Yet with all that outrage, society still seems to function.
Where things break down is in the politics - the governance of our country. The ferocious vocality within the echo-chambers, and the adoption of outrage by even the mainstream media, is leading to the seemingly most popular (or least unpopular) actions - opposing whatever is suggested.
Contrary to what the outrage that will generate the clicks on the article, it is not the end of times. Eventually, the system will balance itself out. Here are some things to consider helping that process along:
· I hope the governments will do what they do best in this matter - absolutely nothing. Governments are ill-equipped and totally incompetent to see the long-term effects of any intervention they will put in place. The matter is complex, and therefore intended regulation will need to be even more complex. And as always, complexity fails.
· The AI's will need a broader set of goals, a larger frame of reference. I would think this is a natural progression of the learning as if the division grows wider, and instils more violent behaviour, it does not increase consumption of content. To put it bluntly, it is difficult to consume content when your head is being bashed in, or while you are bashing someone's head. A milder way of putting it is that civil unrest escalated into violence is simply bad for business, there is simply less money for advertisement going around in times of crisis.
· Let reason prevail over outrage - this is something we all have a responsibility for. It is up to us to be in an echo-chamber or not. It is up to us to only consume the news that has been curated for us, or to consume news that we don't want to hear. Facts don't care about how you feel about them, they remain facts all the same - so the world will not end if you read or hear something you fundamentally disagree with. Take back the control and realise it is your choice to be outraged, it is your choice to reinforce the AI with the notion that this is what you want.
Evolution of AI
The evolution of AI will continue. There will be challenges and there will be opportunities. I personally believe it is true when we refer to this period as the "fourth industrial revolution". With industrial revolutions come changes in the social construct. Whether we like it, or we don't, it simply will happen. The singularity is often referred to as the tipping point. The point where the AI is indistinguishable from humans in form of intellect and creativity. Whilst I think this is a great achievement, I don't think it is has the potential for disastrous impact as a lot of people seem to fear, The AI has a purpose that we give it. When the AI becomes self-aware, that is the point where I would seriously worry about the next evolutionary stage. The fact that the Facebook bots invented a new language to communicate together in a more efficient way does not keep me up at night - it found the best way to achieve a purpose as it was given. I would be concerned with an AI that sets its own purpose, irrespective of its human creators. I don't think we are close to that point, and AI right now is a tool at our disposal - it will learn to achieve what we tell it to.
Originally posted: 17 August 2017