In Hollywood, the vision of Artificial Intelligence includes Skynet controlling our nuclear arsenal. Robots in ‘I, Robot’ designed to help us in our daily lives. Or Ultron from the Avengers as “A Suit of Armour Around the World”. The theme for you movie lovers out there will be clear in these examples. AI is bad and machines will be the end of the human race.
However, while it’s always good to be wary of the practical uses of AI, the reality is very different from what’s usually portrayed on the big screen
In reality, we use personal assistants and Smart Hub tech like Siri and Alexa to answer our questions and help organise our day. AI built into platforms like Netflix and Spotify give great movie and music suggestions based on what we’ve seen and heard before. And companies like Amazon use AI to sift through your purchase history to make intelligent recommendations for you to ‘add to your basket’.
So What Is It…Really..?
In simple terms, it’s the demonstration by machines of a level of human-like intelligence. Simple right? Some of the human-like indicators of intelligence that AI can attempt to mimic are:
- Planning
- Learning
- Reasoning
- Problem-solving
- Knowledge representation
- Perception
- Motion
- Manipulation (of objects)
The easy way to figure out if something is AI or not is to ask yourself, ‘Is this bit of tech mimicking any of the human indicators of intelligence?’. If the answer is yes, then it’s probably AI.
What about Machine Learning?
Machine Learning, or ML, and AI get mistakenly referred to as separate thing in fact, ML is a type of AI. It’s the mimicking of the ‘Learning’ aspect of human intelligence. It’s where a computer system or program uses huge amounts of data to learn how to carry out a task.
In human terms, reading hundreds of French-language books to try and learn French would be an example of human learning. In computing, software that takes in gigabytes of financial data in an attempt to learn where a share price will be at some point in the future is a good example of Machine Learning. So Machine Learning IS AI!
Where Did It All Start?
We can trace the first practical use of AI back to what some consider the first physical computer, created back in 1939 in Bletchley Park. It was here that Alan Turing, along with his Mathematics compadres, designed ‘The Bombe’, a machine that cracked ciphers created by the Germans during WW2.
Below is a scene from The Imitation Game. It shows the Bombe in its full glory, cracking the Enigma encryption and saving thousands of lives in the process:
In this case, the human-like intelligence the Bombe replicating was ‘Problem Solving’.
After his triumph, Turing produced detail designs for the Automatic Computing Engine (ACE) whilst working for the National Physical Laboratory. These turned out to be the designs for the modern digital computer as we know it today with processing capability as well as a built-in memory.
And What About AI Today?
So we’ve talked about the first use of AI but what about current practical uses? And as great as Spotify, Netflix and Amazon suggestions are, let’s be honest, they aren’t exactly the most interesting examples of AI.
Automobiles
Although people still think of driverless cars as a thing of the future, Elon Musk has brought this to the forefront of peoples minds. And directly into peoples driveways. Tesla has changed the game and it leads the way when it comes to electric cars and especially self-driving vehicles. Not only is it very very cool, but it also gives back to humans something that has been traditionally impossible to recoup. Time. Imagine driving to a business meeting and being able to finish a presentation off while sitting in the driver’s seat?! Well, now you can. Although, the boys in blue will probably pull you over because it’s just not fully normalised…yet.
s, loans and direct debits. But it also gives a greater sense of legitimacy and validation to the challenger. Revolut’s page announcing when they obtained a banking licence shows what it meant to them.
Financial Services
Fraud – This has been a big issue for financial institutes and consumers but AI has helped cull the number of incidents. It does this by analysing your purchasing history and understanding whether a particular transaction is ‘out of the ordinary’ or not. It then sends a message asking to confirm whether you initiated the transaction or not. All without the involvement of a person on the other end.
Credit Cards – It’s not only fraud where AI has positively affected Financial Services. Companies like American Express purchase vast amounts of data about various individuals purchase history and creditworthiness. It gets fed into their AI system to figure out who’d be the best candidates for their fancy new credit card. And finally they send these individuals direct promotions about the new service in the hope that they’ll sign up. Again, all without the need for a person on the end of a phone or typing on a computer.
Robo Advisors – Robot Financial Advisors are another great example of AI in FS. These are like for like replacements for the traditional, 50-year-old ‘man in a suit’ that would traditionally be the person giving you financial advice. Now, you can give an AI application answers to a few questions and within minutes you could be investing in something that suits your financial needs for a lower cost. Companies like WealthSimple and Nutmeg are two of the more popular platforms in this area.
NB: I’m not sponsored by either WealthSimple or Nutmeg but happy to take a huge sponsorship deal if either of you wants to reach out with buckets of cash!!
Customer Services
The Customer Services sector has had a very subtle introduction to the AI space. Many companies use sophisticated chat software to replace customer support functions (with an actual human as a backup). However, this is just the beginning as speech recognition and human voice replication software is improving at a rapid pace. Ever been fooled by a call that was an automated voice trying to give you compensation for an accident you weren’t in? I have. In fairness, it was only the first 10 seconds I was fooled by. Because when they started talking over me I realised it couldn’t be a human British customer services advisor. Because a Brit just wouldn’t do that.
Sport
This is another area where the implementation looks subtle. Formula 1 is the sport where tech has been most prevalent and AI is part of that. When Lewis Hamilton is driving laps, Machine Learning is employed to churn through gigabytes of data received from the car. The team then uses the data to tweak how Lewis should drive the car or what tyres to use. The team then relays the instructions back to Lewis who then acts on it…..or not in some cases.
Biased Computing
It’s not all sunshine and rainbows when it comes to Artificial Intelligence and its practical applications. Apart from the obvious doom and gloom of AI controlling nuclear weapons and seeing humans as the enemy of the planet, there are some issues with AI that are closer to home. AI Bias is one of them.
The reality of AI is that it’s created by humans to mimic human intelligence even though it’s seen as somewhat of an organic entity. Or to be more specific, it’s mainly created and programmed by white males and as such, comes from their perspective.
So am I just adding to the list of people just bashing white males? In short, no. The premise of Artificial Intelligence Bias is entirely based on logic and facts.
In terms of education, white males currently dominate the demographic of people taking computer science as a major(see below). The next closest group is white women with a huge gap in between.
This lack of diversity flows into the industry itself. Males dominate the industry, making up over 75% a 2017 study showed.
Just to be clear, when it comes to bias, it’s mostly unintentional.
Remember when Apple launched their famous wearable tech watch with a cool built-in heart rate monitor? Well, it transpired that the infrared light didn’t work on darker skin making the feature useless for this group. But people didn’t assume that it was an intentional design decision to exclude the black community. It was more a result of the demographic of the people who were in the room, designing and testing the product, and the lack of representation RATHER than a racist conspiracy.
If all the people testing and designing the product were white (and the evidence points to that) then the product will be perfect for a white person and less so for a non-white person. It’s that logical.
Now, the Apple watch example isn’t really AI because it’s not really mimicking human intelligence. However, there are better examples of AI Bias that have the same lack of representation leading to a biased outcome.
Joy Buolamwini, a scientist from MIT, published research uncovering racial and gender bias in facial recognition software sold by big tech firms like Microsoft, Amazon and IBM. When trying to guess gender using a face, research revealed that the software performed better on male than female faces. And when it came to error rates, the rates of error were less than 1% when identifying light-skinned males and a shocking 35% when it came to darker skinned women.
Here’s the summary video that’s also on her site, GenderShades.org:
When factoring in China’s social credit system the accuracy of AI and facial recognition software becomes even more critical.
So it’s pretty clear that we need to address existing AI bias before we all give in and the government replaces all jobs with robots and AI. Here are some more bite-sized examples of AI bias that you can tell your friends about and look smart and cool in the process:
- A driverless car is 5% less likely to spot a dark-skinned pedestrian than a lighter skinned one
- Most products using infrared detection like soap dispensers and hand dryers have a hard time detecting dark skin, as dark skin reflects infrared less easily
- Amazon scrapped an AI recruiting tool it built after it heavily discriminated against CVs that had the words ‘women’ or ‘woman’ in it. This was because the data used to build the tool was pulled from mens CVs which hardly contained either worA driverless car is 5% less likely to spot a dark-skinned pedestrian than a lighter skinned one
To summarise, the bias doesn’t come from the tech itself, it’s comes from the people who are building it. So to tackle the problem of AI Bias, make the workforce AND the data used to drive the decisions and testing more diverse. It’s pretty simple.
So…What’s Next For AI?
There are some applications of AI that have already been covered and while I’ll try to tread new ground, a lot of the near future of AI will actually be a significant supercharging of the AI tools we use today.
Driverless Cars
So this is the first instance where I’m covering familiar ground. Although Tesla has made a good start with driverless cars, future AI will ensure all cars are driverless and to a certain degree, connected. Every car being part of a network, run by efficient AI, means removing the element that causes most accidents. Human interaction. And if AI connects all the cars on the motorway it means that speed can be more efficiently regulated leading to less traffic, higher average speed and better fuel efficiency. Like a Blockchain of cars!!
Home Assistant Robots
This is again walking over trodden ground as we have home assistants like Alexa and Google Home. But the future will be a lot more sophisticated than a voice in a speaker. There are already a number of other home appliances that use a degree of AI. There is, for example, a washing machine you can load that will track your daily habits and start a wash cycle that finishes as you get home. And a fridge that reminds you that if you’re planning on having your regular ‘Taco Tuesday’ that you need to buy another pack of chicken and some avocados.
But as you might have seen, Boston Dynamics is making serious and literal leaps when it comes to humanoids. NB. Check out their homepage. It’s truly scary amazing! The incredible leaps they and others have made could see us having a humanoid helper do our washing up, sort our laundry, take delivery of packages, clean and generally do all the things that we’re too busy or too important to do.
Speech and language recognition
Although some automated customer service lines are reasonably realistic, we’re not too far away from a world where AI runs ALL customer service calls and are indistinguishable from a human response. With some recorded voice customer lines and some timed or based on keypad entry we’re in a kind of limbo state.
Healthcare
This is an area where tangible examples in our daily lives are surprisingly limited. This is, in part, due to the regulatory constraints around health compared to other industries, but also concern from consumers. There is something about a doctor or a surgeon that gives a certain level of reassurance that, at present, a robot with an AI brain just can’t give. Despite a machine having better consistency when it comes to procedural work. However, once we’ve jumped the regulatory and consumer confidence hurdles this will be the area AI hits hard.
Art
Creative industries are the hardest for machines to mimic because creativity is quite difficult to measure. When it comes to REPLICATING a piece of art, AI is relatively good. Mainly because it can analyse the colours on a canvas and the exact positioning of brushstrokes and pixels. But when it comes to creating an original that’s acclaimed by critics, it becomes tricky. However, as Machine Learning has proven, it just takes the right amount of good quality data for a machine to, firstly, understand what good art is, and secondly, create an original piece based on what’s popular. This applies equally for most creative industries.
It’s not necessarily a direct response to challengers but the dramatic spate of redundancies could be partly down to challengers demonstrating that building and running a bank doesn’t require millions of employees. HSBC is the latest to announce redundancies after Deutsche, Citi and Barclays. Whilst these redundancies aren’t specifically in their retail banking division it’s a clear indicator that all banks are looking to lean down to reduce overheads.
Summary
So hopefully you’ve learnt that AI isn’t that scary after all. Partially because we’re actually not as close as people would have you think to an AI induced nuclear destruction. And also because it’s us humans who have our own bias and imperfections who program the Artificial Intelligence. The other issue is that the hardware that runs it is still limited by size and performance. It’s something that will change with the rise of Quantum Computing but that’s too complicated a subject for this article. It has the word ‘Quantum’ in it after all. In short, Quantum Computing will make creating AI infused machines quicker and smaller and smarter. And it’s in fact, the only feasible way to make a machine think the same way as thehuman brain.
But we’re still quite a way off that right now. Or are we. Did a human write this blog or was it a machine…..
