Introduction

I had a great, quick catch-up with my old mate Nayur Khan the other day. Over the last decade, Nayur has been working at the bleeding edge of Artificial Intelligence, Machine Learning, and Data Science, delivering commercial projects. He’s a bit of a subject matter expert in this area, so much so that he’s been recognized in the DataIQ 100 – a list of the top 100 influential people in data.

We covered a few different areas during our conversation, but the key point was to shine a light on the real position of Artificial Intelligence, which may be different from the view that has been created by social media.

NOTE: all images below were generated through AI using DALL-E

What is the realistic position of Data Science, Artificial Intelligence, and Machine Learning?

Since the release of ChatGPT in November 2022, we’ve witnessed a global phenomenon in interest for Artificial Intelligence. As Nayur said, “AI has been around for a while, but the last 6 or 7 months has really captivated an audience, moving AI from research to the mainstream public. An AI tool, ChatGPT reached 100 million users in about 3 months or so. We haven’t seen this kind of user adoption before in such a short span of time – this has led to an explosion of innovation that has started a flywheel effect of competition between a lot of organisations”. In fact, ChatGPT is the fastest-growing consumer application in history.

For many people across the globe, ChatGPT was the beginning of Artificial Intelligence. In reality, however, Artificial Intelligence has been around for many years and has been quietly slipping into our day-to-day lives. Nayur pointed out that “it suddenly feels like we have moved from having the somewhat Neanderthal like technology within our homes, that you used to ask to turn on the lights or turn off the lights, to really good conversational type AI. It feels very surreal sometimes, and definitely like we made a jump”. The jump is as a result of one of the global tech giants making such a public, step-change release in the AI space, ChatGPT. It has captured our hearts and imagination. In what felt like an instant, AI was propelled from being predominantly in a phase of research, into the public domain.

Allowing a virtual assistant into your home to turn the lights on and off is a tiny, comfortable step forwards – although there have been questions about eavesdropping, ethics, and how invasive this use case is. In many respects though, ChatGPT does indeed feel very surreal as it appears so human in the way it interprets language and communicates responses. It generates deep, seemingly insightful answers to questions, in timescales that surpass human capability. Some newcomers to AI, drawn in by the ChatGPT hype, feel that these seemingly superhuman capabilities are already too powerful and are calling for a halt to development. But in reality, we really are still in a research phase. There’s a very visible indication of this in the user interface of ChatGPT where it notes it:

• May occasionally generate incorrect information
• May occasionally produce harmful instructions or biased content
• Has limited knowledge of world and events after 2021

ChatGPT has been placed on a pedestal and this has provoked a very visible fight amongst the tech giants, and a new wave of innovation has already begun. In a few years, expect to see high profile global tech organisations that have only recently been formed as start-ups around this new technology. “There is this whole continuing innovation, which is really difficult to follow or grasp what is going on here” suggested Nayur. “Are we seeing a disruptive change in front of our eyes?” Well, as Bill Gates put it: The Age of AI Has Begun.

The isn’t the first large scale rapid innovation we’ve undertaken. The industrial revolution (approx. 1760-1840) saw rise to mills, industrialisation, mechanisation, steam power, chemical manufacturing, gas lighting, mining, cement, iron industry, glass making. The term Luddite was coined for people who protested against manufacturers who used machines in ‘a fraudulent and deceitful manner’ to replace the skilled labour of workers and drive down wages. However, the release of ChatGPT is slightly different, as Nayur continued, “this feels like another explosion, it just feels like a larger scale because millions of people have access to the Internet, millions of people have access to mobile devices and now they have got access to this incredibly interesting technology”.

The flywheel effect Nayur mentioned can be seen across social media too, as people realize that AI is likely to be a new revolution, and that they themselves could potentially innovate in this space and become a future global tech giant and billionaire. “Whenever there’s a disruptive kind of element”, Nayur began, “you’ll have lots of excitement, lots of people giving their opinions and thoughts, even though they might not have been in the field that long. Some of this information will be good and some may not be”. The Dunning-Kruger effect, a cognitive bias where individuals with limited knowledge or competence in a particular area tend to overestimate their abilities, is in full effect. Content creators and commentators mistakenly believe they are experts in AI when, in reality, they have only a superficial understanding from interacting with ChatGPT. This leads to a lot of information and misinformation, just as we saw during the Internet boom, the release of the iPhone, the advent of computing, and the industrial revolution.

Changing the text slightly to ‘draw a bird, make it look metallic, give it a robotic theme in a vincent van gogh style’ will generate a different output. This enables individuals with limited or even no creative skills to request their masterpieces. The whole basis of art is being challenged and a new genre of AI Art is emerging. In this space, Generative AI is causing us to rethink art itself. Similar challenges are appearing in music, film, academia, research, journalism, etc., anywhere that boilerplate content can be created but, for the time being, AI is unable to achieve anything like genuine human creativity and abstract thinking – like Van Gogh.

Nayur posed a great question about Generative AI, “everyone’s feeling, ooh, this is quite cool, this is really, really quite novel. Here’s the question many are asking…what can we solve with this tech?”. Whilst Generative AI is cool, and intriguing, and it is always fun to add a bit of a human creativity spin to the output, there are other AI technologies that are already having a meaningful impact. Further advances are on the near horizon too. The Gartner Hype Cycle for AI [July 2022] provides a sensible view of the current state of AI. It suggests, for example, that despite the surrounding hype, autonomous vehicles are in a research phase and expected to arrive in around 10 years. Even Generative AI is illustrated in a research phase and expected to arrive properly in 2-5 years.

Nayur notes, “I think this is where we are right now”, he continued, “with this flywheel effect. Lots of different things are popping up (weekly in some cases) – then you’re left scratching your head wondering what are the real problems we can solve – will this tech help?”.  One thing is clear though, it feels like the tech is driving the innovation. This is the opposite of the Internet and iPhone, where the innovation was a response to real world opportunities.

What Commercial Applications are you seeing for Artificial Intelligence?

We’ve always had challenges where vast amounts of data can provide valuable insights. Previously engineers would have coded algorithms to tried to understand and interpret the data, which as anyone who has tried this will attest, is difficult! What is new is that Data Science has led us to building models and algorithms that interpret them. These algorithms can then make predictions based on the models. Some of these algorithms and models are already quite well understood and defined.

Whilst Generative AI is intriguing, there are already real, commercial applications of Artificial Intelligence. We’ve been exposed to some of these for years. “It’s not new for us as consumers. The recommendations for films we should watch, the music we listen to”, began Nayur, “Google Maps is an example that I will  pick on – since I rely on it so much! There’s AI involved in our lives whether we realise it or not. But they’ve always been for a particular purpose”.

On the Gartner Hype Cycle, the only item that is close to productivity is Computer Vision – the ability for technology to ‘see’ and interpret the real world. Computer Vision is an area of AI that is already having an impact on our daily lives. Examples of Computer Vision include: automated numberplate recognition, handwriting recognition, facial recognition (see later for ethical challenges with facial recognition), and also pattern recognition for voice and language. These should all resonate with day-to-day tasks we complete like parking our cars, passing through Passport Control, person recognition from home security cameras, and speaking to virtual assistants.

“Some of the businesses have also been using AI over a number of years”, he continued, “maybe not as much as they could have, but”:

• “In telecommunications, predicting network degradation or improving quality of experience of telephone or mobile lines of 4G and 5G” – what is AI for networking
• “In healthcare … for detecting cancer” – computer Vision has been prototyped and well received within healthcare settings
• “Creating new vaccines and molecules used to take us decades and now were down to a fraction of that time” – AI is supporting and “accelerating clinical trials so we can speed through the process”
• “Optimizing manufacturing or reimagining complex supply chains” – many applications
• “In finance, we’ve always had ways of detecting fraud without AI but now with the amounts of transactions that happen, one of the great things about the technology is being able to look at such vast amounts of information and detect that in real-time and be able to do something about it” – fraud and anti-money laundering

“AI across a number of different industries has so many different good use cases”, said Nayur, “but regardless of what we have been doing for the last few years, and what we are seeing today, I’ve always loved what Steve Jobs said when he talked about tech, what are the problems we are trying to solve? Why is that a problem? Then work your way all the way backwards to the tech. That recipe still holds today”.

There is a key reason why researchers identify AI as being in a research phase, perhaps one of the reasons why businesses haven’t been using AI as much as they could, and why the Gartner hype cycle indicates it will be years until many of the technologies achieve general productivity, is an underlying challenge. Ethics.

Artificial Intelligence and Ethics

Ethics means “way of living” and encompasses our moral principles, how we conduct ourselves, and what we believe is fair and just. There are many challenges around ethics within Artificial Intelligence, primarily because machines are unable to make ethical decisions and often have integral biases from the data they consume and the teams and organisations that create them.

Recently, we’ve seen the pessimists asking for a pause to AI development whilst we look at the ethical challenges. Are they right to request this pause? Are we facing new ethical issues as our understanding grows, or have these issues always been present and we’ve just not tackled them thus far? Perhaps we have been too focussed on developing, moving forwards, innovating, rather than resolving ethical challenges. Or, perhaps these pessimistic individuals are modern Luddites and it is just normal human behaviour and history repeating itself. Should we be taking more care to answer these ethical challenges? The industrial revolution had many negative impacts: poor working conditions, environmental damage, overcrowding and poor living conditions, economic inequality, social displacement. Many of the social and environmental issues it created are still being addressed today.  Could any of these issues been avoided or minimised? Are there looming  negative impacts of AI that could be prevented?

“AI has been around for a while” said Nayur, “we have always had a problem”. Nayur went on to identify some specific examples where AI has had inherent issues:

• “Would it be OK to award someone higher grades if they went to private school, but downgrade them if they went to public school?”, asked Nayur. “An algorithm in this country (UK) that did exactly that and it penalised a lot of A-level students and GCSE students when they couldn’t go to school (because of Covid)”. This has a lasting impact on students during Covid years
• Facial recognition systems have consistently failed to work with non-white skin colours. “There is a test case right now in the UK where driver unions are taking action because a number of their darker-skinned drivers were penalised because they could not be identified by the facial recognition software – they unfairly lost their jobs”

“I don’t want people to lose focus that there are actually problems today with AI, and we should be tackling them today”, Nayur continued. “To me it is a serious thing … for the A-level students, I saw first hand the effects. There is no apology, when all these things happen.  It is not very transparent why the algorithms made the decisions they did.  It is not explainable, and no one takes accountability. If we want to build technology and use it for a better future, we’ve got to take a step back and acknowledge that there are systematic biases in the past that are encoded within the data. There are unconscious biases that we have, and unless you address all these things up-front, there is a high potential that you will marginalise and discriminate against some people”.

Many of us are already aware of quite public ethical challenges around AI:

• ChatGPT has been accused of having hallucinations and numerous occurrences of biased data have been observed
• We’ve already mentioned above about virtual assistants in our homes. Whilst we get benefits such as effortlessly playing music, quickly getting answers to questions, or setting reminders, there have been many questions about eavesdropping. An always active microphone means supposedly private conversations can be heard, potentially recorded, and even listened to by other humans

Perhaps we need to stop accepting, and stop blaming the system, and take a step backwards. Upfront thought and design are needed to ensure that biases aren’t inadvertently introduced into solutions and this approach is being considered in some areas.

In the UK, the Financial services sector knows that AI can provide a lots of benefits. As we mentioned previously, AI is an excellent solution for detecting fraudulent transactions or attempts by organised crime to launder money. The Financial Conduct Authority (FCA) and Financial Service Authority (FSA) also understand that if AI makes incorrect decisions, individuals or organisations can lose money. To this end, these governing bodies do not currently allow Artificial Intelligence to make decisions outright. Instead, AI is able to make suggestions to a human; the human makes the decisions, but they can also consider the view from AI. Whilst making the decisions, they are also able to provide feedback and train the model. This kind of augmented approach, or assistive AI is not there to replace humans but enhance decision making.

Similar approaches have been seen in cancer screening where, in the UK, two radiologists currently review all scans, and potentially a third radiologist if there is a disagreement. This is time consuming but designed to reduce the chance of human errors. A human-AI augmented view has been trialed in the UK and has reduced false positives by 1.2% and false negatives by 2.7%. In the US the humans supported by AI produced 5.7% fewer false positives and 9.4% fewer false negatives. Ethically, this solution approach is sound as there is plenty of clean, un-biased data and the system is being trained correctly before being allowed to make automated decisions that could literally impact people’s lives.

“These new Generative AI tools and technologies are not to replace but to augment”, commented Nayur. “The simple example is Google Maps, navigation. It’s not there to tell me I must take a route; it suggests several potential routes … it is up to me whether I take those routes or not.” Nayur continued, “It works so well in this kind of co-pilot type mode, and if we start looking at co-pilot type technologies that augment us, that help us with some of the mundane tasks that we do, that’s a fantastic way of looking it”.

This augmented or co-pilot approach to AI is gaining prevalence and will likely be the how much AI is introduced over the coming years. AI won’t be replacing jobs but supporting, whilst hopefully increasing accuracy and reducing mundane tasks. We’ve seen a shrewd move from Microsoft, one of the key investors in OpenAI as they have introduced ChatGPT based co-pilots, in this exact direction as into Windows/Office, software development, CRM and ERM and more. Whilst co-pilots don’t resolve ethical issues outright, they are a small step forwards that provides benefits for humans, whilst as an industry we hopefully look at some if these inherent issues.

What advice would you give to somebody looking to move into AI?

With so much hype around the industry and the technology, and opportunities to be part of the next wave of innovation, it is natural for people to want to be on the journey. “It is an incredibly wide field”, Nayur began, “it takes a football team of skills to get involved and you can bring your existing IT, software engineering skills, or DevOps skills and start developing Data Science skills or analytics skills”.

There are the more obvious AI roles around the Data Science: the maths; the statistical methods; the data; building the algorithms and models. But there is also a need to operationalise the models, to monitor and maintain them, “which is slightly different from monitoring infrastructure”, said Nayur. “You train models for a point in time and after a while you need to come back because the data they were trained on is out of date. As an example, models that were trained based on customer behaviour, were trained in a certain way … when Covid hit, pretty much all those models stopped working. Why, because we were all in lockdown and customer behaviour had completely changed. You have this new problem, and new emerging types of problems that you need to detect where data has drifted, or things have drifted in the world, and so you need to update these models.”

And, as discussed above, ethics needs to be considered as part of AI projects. There are new opportunities for anyone that is interested in contributing to this area.

But perhaps the key to success, which again is similar to existing technology projects, is the role of a product manager. A product manager should be focussing entirely on identifying the problems and the opportunities that AI may be a solution for, not looking at the technology and trying to find uses. Nayur makes the point that, “One of the things we miss a lot … is that product manager mindset; I’m not just building a model for the sake of a model, I’m building a product, something that people are going to use”.  The product is the solution, not just the cool technology that underpins it. In the case of ChatGPT, for example, “the model is one part of it, but it is actually the interface that is quite cool, you just type in some things and it has a got a very good feedback. It is what people interact with that is going to hook people”. Nayur makes a really good point here. Users traditionally hate “talking” to chatbots. One of the first things they do on a webpage is minimize/kill the chatbot that asks if they need any help. Whilst the AI that powers ChatGPT is amazing, it would be nothing if the UI didn’t write out the response as if a human were typing it. It is the product as a whole that has captured our hearts and minds.

There are numerous opportunities to get involved in this fascinating, innovative space. It really isn’t too late. There are new, emerging roles but there are plenty of opportunities to bring your existing skills.

Final Thoughts

Despite the hype, be mindful that AI, including ChatGPT and Generative AI, is still in its infancy. Over the coming years we are going to be spectators to a very public fight between the global tech giants. We are going to see a whole new wave of innovation. There are companies that don’t currently exist that will be global names within a decade. And regardless of whether you are an optimist, a realist, or a pessimist, our lives are going to change forever. But, similar to when fear existed during the Internet boom, mobile phones, the computer, and the industrial revolution, new jobs will be created around these new innovations that will mean humans aren’t just replaced.

And finally, there are significant ethical challenges that need to be addressed before AI can really be trusted to act on behalf of humans, and we should all be pushing the tech companies to make sure there isn’t a shortcut in this step. As for Terminators ………….. we’re a long way from that, yet.

More AI articles