🍁🤖O Canada. Artificial Intelligence’s cradle and the death of the musician?

This week looks behind the scenes of artificial intelligence, gaps today and changes already in motion.

The AI Winter

Early AI (The Mark I Perceptron)
(via Wikipedia/Cornell University Library website)

Tucked away in the depths of universities are projects that have little love in the wider world and yet can lead to a significant breakthrough.

For decades, experts were down on the idea of neural networks but over in Canada, a small group of expertise grew regardless. These approaches led to the advances in machine learning and the now very popular area of artificial intelligence.

Toronto Life has an excellent article on Geoffrey Hinton, one of these early pioneers and describes the “AI Winter” when few believed.

One of the reasons for persevering with neural networks was due to our lack of understanding of how the brain worked. Hinton said:

“The brain has got to work somehow and it sure as hell doesn’t work by someone writing programs and sticking them in there,”

The article also looks into his history and he has a fantastic family tree ranging from the founder of Boolean Logic to Everest and the Manhattan Project. He also happens to come from Wimbledon 😊

More here.

Artificial Intelligence’s holes

How far we have come with artificial intelligence should be celebrated but there are still holes. Wired takes a look at the downsides finding it:

  • cannot think abstractly;
  • does not scale to become human-like;
  • useless without large data sets.

One of the benefits of this breakthrough though is the billions of pounds being spent finding other techniques that improve on these drawbacks. Without it, the AI winter would have continued.

Changing workforces

The unknown can create angst and the potential of AI and robotics has many worrying about jobs (though automating away manual repetitive tasks can only be a good thing for people if we get it right).

Wired looks at how robots have entered the workplace as painters and changed the jobs of the people around them.  The robots took over brute sanding and painting, while humans did the more complicated tasks like assembly, with some looking after the robots.

Reuters takes a look at how eastern Europe is also automating its factories as the cost of labour increases and the number of workers decreases.

In the US, a lack of labour is resulting in cows being increasingly milked by robots. A study by University of Minnesota researchers shows that the economic benefits are not uniform though. Robots were more profitable with upto 240 cows but at larger numbers (they studied 1,500 cows) parlours remained more profitable. More here.

Finally Newsweek has a rather alarmist article about the end of the musician based on Spotify hiring an AI scientist. 

Spotify has an army of people working on the huge swathes of data available to it and when I see them at conferences, what they achieve never fails to astound.

Spotify and indeed IBM say their focus is on assisting the musician rather than replacing them, which holds true today. If the AI cannot interpret quality, which of course it cannot with current techniques, all it can do is generate millions of tracks in the hope of creating a gem. 

Would you trawl through the results?

💀💀The end of advertising?

This week – attacking advertising seems to have bubbled up to the fore so I look at two leading commentators perspectives and why I think they well – miss the point.

Finding the villain.

Advertising has been the villain for as long as I can remember. Even internet marketers found a way to criticise TV advertising because you couldn’t measure it.

Farhad Manjoo in the NY Times has a piece on advertising being the central villain online.

He suggests that without it many issues we are seeing online today would disappear, giving two examples:

  1. Russian trolls would not be able to hone their messaging as hey could not test its impact using advertising first.
  2. “Nutty content” on YouTube Kids (I assume he means the computer generated animations I discussed in a previous newsletter).

Finally he suggests that the ad industry produces endless incentives for gaming the system that are only fixed after they appear.

It is an interesting proposition and certainly the world would be a different place without advertising. It would result in some very well known companies disappearing, including potentially the very publisher Farhad is writing for.

Overall though, it would be better in some ways and worse in others. It is difficult, maybe impossible, to get consensus on or even understand the full impact. I think it would limit innovation online for example. But there is one thing I am sure about. There will always be people looking to game a system. The system would just be different.

Actually, I could argue the true villain in both those examples is artificial intelligence as they are really about easier to use interfaces built on top of algorithms that make content faster to create and easier to target. Just this week, the creation of “deep fakes” became mainstream thanks to increasing awareness of software that allows you to change the face of someone in a video.

But I am not arguing that either.

Rapid innovation is the true villain, as everyone rushes to adopt new approaches in a positive way, others rush to take advantage in, let’s say, less positive ways.

I don’t believe anyone would argue for a slowdown in innovation, though those affected by it might like to see it slowed down. In the end, as the innovation matures, regulations appear and there are less loop holes to exploit as the successful companies will be those that fixed the issues.

Don’t blame the media.

Umair Haque also wrote a piece on Medium last week on why the failure of media is not the fault of technology but advertising.

He suggests that the ad agencies took the easy route of reinventing the offline billboards by creating banners and using algorithms to target them rather than a human, creative one.

Apple tried the latter approach though even this was positioned as a reinvention of TV advertising. They eventually, after issues with pricing and getting the ads live, shut it down in 2016.

Publishers could also try these types of approaches – though they are not exactly incentivised to encourage clicks that take their readers away from their website.

With display advertising, the attention should not solely be on clicks but whether it was seen. Of course, people screen out ads but the best ads can cut through – even online – and especially when they make it to the right audience.

Keeping things simple, advertisers will go where their audiences are and then (theoretically!) spend as little money as they can to deliver their message to them. Inertia inevitably happens though.

For as long as I can remember, Mary Meeker has in her annual state of the Internet reports, pointed out that advertising is overspending in print and underspending in mobile. Advertising spend in print is finally declining whilst spend in mobile advertising is indeed rising.

Outside of Google and Facebook, advertising is much more difficult – especially for smaller companies. Upgrading the platforms to make them more accessible will happen over time as the use artificial intelligence gets integrated more and more. Transparency and reducing fraud both remain a heavy focus for the industry today though.

The smallest companies will never be able to invest significant sums of money in creative as Umair would like to see though and if everyone did manage somehow to hit that level then an even higher bar would be required to stand out from the crowd.

🔥🔥Artificial intelligence more profound than electricity or fire.

A deep dive into artificial intelligence this week looking at some of the announcements about what is possible today as well as a look into the future disruption of the workplace.

Possible today.

The possibilities from artificial intelligence are picking up steam again (did it ever really stop?).

Airing on YouTube later today, Google CEO, Sundar Pichai and Youtube CEO Susan Wojcicki are interviewed on the successes and challenges tech is bringing to the workforce today.

We already know one soundbite from Sundar in the interview:

Artificial intelligence will be more profound than electricity or fire.

He also talked about how much we already use artificial intelligence in our day to day lives. Just yesterday, someone was stunned at how easily I could find a photo from a few years ago thanks to Google Photo’s ability to recognise objects in a photo.

Popular definition of artificial intelligence remains something out of a Jetsons cartoon though and this definition has been generations in the making. Rosie the robot, from the aforementioned cartoon first appeared all the way back in 1962. The result is that stories of what artificial intelligence can do in discrete situations get extrapolated beyond what is actually possible.

What is possible today though is still astounding.

Microsoft recently showed that artificial intelligence can imagine a bird from a short piece of text: “Create a bird that is red and white with a very short beak.”

The result is the image below.

There is clearly plenty of imagination going on in that photo – nowhere was the shape of the bird mentioned, nor that it should be placed on a branch.

However, the researchers showed that this imagination came from historical knowledge of vast numbers of photos viewed. More images showed birds on a branch than flying. So really this is more literal imagination and human-like imagination. When the researchers suggested drawing a bus floating on a lake, it struggled.

This still opens up potential new tools bringing new opportunities. This for example might eventually change the meaning of stock photography or allow Photoshop to fix images quickly.

Just last week, Photoshop rolled out an AI tool to select objects within an image – something that was a laborious task in the past.

Some might contend that it cannot do it as well as if it was done manually and that is probably true but it is unlikely to be noticeable in most scenarios.

Even today things go wrong with photo editing with Vanity Fair recently giving Oprah three hands – something a future AI might be able to warn of or prevent completely.

Tomorrow?

Whilst there are plenty of opportunities with artificial intelligence, perhaps the biggest fear is around the workplace with people imagining upheavals on a much larger scale to the disruption of the textile industry in Great Britain as factories came online.

The World Economic Forum have released an excellent report looking at the future of the workplace and how jobs might transition in the future. Download it here.

Some of the highlights:

  • Women are more affected by this disruption than men (57% of jobs disrupted)
  • Government, trade bodies, companies and individuals will all need to work together to minimise disruption
  • Administrative and production roles are expected to see the most upheaval by 2026
  • Something that always gets overlooked is the impact on leadership roles. How will those change? The Harvard Business Review takes a look.

One aspect that is already changing is the ability to make better and faster decisions based on sifting through the vasts amount of data that companies now generate.

Here bias may have an impact on businesses. Artificial Intelligence has been shows to have the biases of their programmers built into them. As companies adopt uniform technology platforms, their ability to make unique decisions that outperform competitors may reduce, making it even more important to understand how decisions were came to.