AI and Machine Learning

AI and machine learning: the story so far

In April 2019 we wrote an article, will AI and machine learning take over the world? That year’s South By South West had a programme heavy with talks and conversations on AI and machine learning.

Now, in 2023 it seems the world is asking that very same question again.

Over the last few months, various AI programmes have made international headlines, for varying reasons. In this article, we will ask, in light of recent developments and news around AI and machine learning, what is the current state of play around artificial intelligence. We will look at how and why ChatGPT passed an MBA, and analyse the AI images that have fooled photography experts. We will explore how Microsoft could potentially rise from the ashes of Google and how the latter made a comms disaster of their own AI launch.

But first, let us address and explain some of the recent advances in machine learning and artificial intelligence.

Natural Language Processing (NLP)

No doubt you’ve heard the name ChatGPT by now. It has been making massive waves within the international media and is currently winning the natural language processing race.

ChatGPT has the capacity to take a series of written prompts and indicators and create comprehensible human language as a result. We have seen the headlines of ChatGPT’s copy output being used to pass high-level examinations. There is industry feedback about how mass adoption of this technology could damage the prospects of working writers, copywriters and many other professions which have creativity at their hearts.

Stable Diffusion

Stable Diffusion is a deep learning, text-to-image model. With the release of Stable Diffusion in 2022, text-to-image AI has come on in leaps and bounds. Other available platforms include Dall-E and Midjourney.

Midjourney is an independent research lab that uses Discord and Stable Diffusion to create images from textual descriptions, known as prompts. Like Chat GPT, the user prompts the tool, and the more refined the prompts the more refined the outcome.

The response so far

The general response to Midjourney and ChatGPT has been rather illuminating. At face value, it certainly seems that the wider public and industry response to ChatGPT has been far more critical than that of Midjourney.

The wider criticism directed towards the likes of Midjourney burned out much quicker than it has with ChatGPT. This could indicate a somewhat cynical view towards the value and importance of creative and design-centred roles.

Putting this aside, the developments around AI and machine learning over the last six to eight months have been significant. So much so, that it has a lot of big businesses scrambling to compete.

Pretenders to the throne

In the last few weeks, some of tech’s biggest hitters come out with their own response and competitor products, particularly with ChatGPT.

It would appear that the Alphabet Group in particular sees ChatGPT as serious competition, perhaps even a problem.

In 2022, Google took the stance that AI-generated copy was against its guidelines. Google’s Search Advocate John Mueller took the stance that AI writing tools should be considered spam. Right around the time that Microsoft purchased ChatGPT.

In late December 2022, Google changed this stance and Mueller took to Mastodon to outline this change of opinion.

Google has seen that AI-generated text was fast becoming a widely used tool, and it couldn’t be seen to be punishing those using it.

Another way of looking at this flip-flop is that in 2022 Google was working behind the scenes to launch its own competitor to ChatGPT. It was sowing the seeds for the release of its own marketable text generation AI, Bard.

If you can’t beat them, join them.

Bard and the perils of machine learning

In February 2023, Alphabet Group unveiled Bard and they learned the hard way that deep learning, AI and machine learning are still flawed.

The launch campaign for Bard saw the machine learning tool give a response to the question, “What new discoveries from the James Webb Space Telescope can I tell my nine-year-old about?

Bard responded that the James Webb telescope took the first pictures of exoplanets, meaning planets outside of earth’s solar system.

Only it didn’t. Bard was wrong.

NASA was quick to point out the error, clarifying that the first-ever images of exoplanets had been taken almost 20 years prior. In 2022, the James Webb telescope had taken its first pictures of exoplanets, but it wasn’t the first to do it.

This had calamitous repercussions for Alphabet Group as its share price dropped overnight, wiping almost $100 billion dollars from its overall share value.

How can we use machine learning and artificial intelligence?

Google’s misstep in launching Bard teaches us some important lessons about the use of machine learning algorithms.

It shows us that, when it comes to the written word, current platforms are flawed and therefore cannot be relied upon to be 100% factually accurate. Because of this, much of the public outcry around its use and replacing actual human input, are somewhat moot.

This costly mistake shows that AI and machine learning are very much basic tools and the human brain is still needed to make the best use of them.

From a PR and comms perspective, Google has been hoisted by its own petard here. Their launch campaign should have been stringently fact-checked. In Mueller’s Mastodon post, he specifically states that he would strongly discourage blindly following the results of any AI text generation.

But that appears to be exactly what Google has done in launching Bard.

Microsoft takes the lead

It’s now evident that Microsoft has huge plans for ChatGPT. After acquiring OpenAI’s ChatGPT technology last year, the tech giants are now set on rolling out this technology in many ways.

Recently, the firm announced a new version of its search engine Bing, which would be powered by ChatGPT’s machine learning and artificial intelligence. It will also be added to Teams, powering up Microsoft’s premium offering.

This is a huge threat to Alphabet’s Google search.

A changing landscape

Satya Nadella’s Bing announcement came the day after Google launched Bard and has the potential to completely change the way online search works.

For many years now, Google has operated its search function primarily as a revenue driver, rather than a means of providing the user with the most relevant and accurate information.

Google’s search results aren’t necessarily prioritised in terms of the factuality of the response. Rather, it delivers what Google wants people to see based on if businesses have paid through ads or how companies have used SEO strategies to appear in any given search.

Introducing ChatGPT into Bing has the potential to really damage Google’s bottom line. Users will have a different kind of search experience. Nadella himself referred to this launch as the beginning of an arms race between big tech.

With Google’s Bard misstep, Microsoft has taken the lead and it could change everything.

ChatGBT passes an MBA, what does this mean?

This has perhaps been the hottest topic in relation to AI and machine learning. But is it really as controversial as it has been made out to be?

Christian Terwiesch is a professor and expert on innovation management at Wharton, a high-level US school. It was Christian who wrote a paper titled “Would Chat GPT3 Get a Wharton MBA?”

To Christian’s surprise, the answer was a resounding yes. However, if we take a look at this paper, and in particular the prompts he gave, it does raise some questions.

Above is the first prompt that Christian gave the computer system. What he is asking the system to do is to find an inefficiency within an iron ore production plant. It is not surprising that ChatGPT was able to answer this with clarity and accuracy.

The reason why is that Internet of Things (IoT) processes which monitor efficiency within manufacturing have existed for a very long time. These kinds of questions are already being answered within the industry by machine learning.

Christian’s second question was about an annual revenue calculator, which could be answered by any finance software forecasting tool. His third question was again another logistical one, tasking machine learning with finding a bottleneck in a processing and distribution plant.

Perhaps the clarity with which ChatGPT answered the questions could be seen as surprising. But the fact that it was able to answer it should not be seen as anything other than ordinary.

What is extraordinary though is that ChatGPT has, by showing that it can calculate estimated revenues and identify inefficiencies within a production chain, shown its potential to act as an open-source version of IoT machine learning artificial intelligence.

Christian claimed that the machine learning model outperformed many of his students in this particular examination. Perhaps the fact that his students couldn’t answer these questions as well as a machine learning system is more worrying.

AI image wins a photography contest

Another recent story was an AI-generated image winning a photography contest in Australia. Fuelling the argument that AI could replace creatives in their roles.

The image, which was created by the Sydney-based company Absolutely-Ai, managed to fool the judges. The company describes it as the ultimate test of the rapidly evolving technology.

The story was picked up globally and the image itself is rather beautiful. Whoever created it clearly has a great level of photographic skill in order to prompt an AI in such as way as to achieve the outcome.

But there’s a problem with this image, can you spot it?

Image: Absolutely AI

The judges of this competition were clearly not surfers or seafarers. Had they been, they might have picked up on a key detail that the AI has got wrong.

The ripples running behind the wave are going in the wrong direction.

A common trope among AI-generated images is that while they can be beautiful, they’re often not quite right. Image recognition is something which the human eye is automatically attuned to, it’s inbuilt.

At the moment AI image, like the image above, is on the crest of a wave. It’s getting better, far better. But it’s not quite perfect enough to replace us

So what can AI and machine learning actually do for your business, now?

The battle for big tech dominance has yet to trickle down into our day-to-day lives. Right now, this is a battle playing out above our normal comprehension, but this doesn’t mean that AI and machine learning don’t have a place in day-to-day business right now.

ChatGPT can be a handy tool for stimulating thought processes and helping to generate ideas. For example, when writing blogs are marketing copy it can be used as a starting point to help overcome a particular task or challenge.

Writers’ block is certainly a real thing. Rather than spending long periods of time searching the internet, or waiting for inspiration to hit, use a tool that can quickly give your own intelligence and neural networks a quick jump-start. A valuable time-saver.

The human element

What’s important to recognise is to solve a problem, you first need to understand what that problem is. Simply applying artificial intelligence to something you don’t understand is not the best course of action.

When using this information, we should not treat it as gospel, nor should we pass it off as our own original work.

Instead, we should treat this as semi-structured data. Approach it with the mentality that machine learning and artificial intelligence have opened the door to completing potentially complex tasks or solving complex problems.

But it has not completed the task for us. This still remains within the grounds of human intelligence and human reasoning to do so.

Using these tools more, and learning how to use these tools to maximum effect creates reinforcement learning for the intelligent systems themselves. In layman’s terms, poor prompts and inputs will result in poor-quality outputs.

Likewise, the more skilled the user is at prompting the intelligence system, the greater value they will receive from it.

Right now AI and machine learning are just ‘tools’ that are controlled by humans. Its implementation depends on the intent of the human controlling it.

The issues facing the AI and machine learning model

One problem is that the data can be biased, skewed or even incorrect. When that data is left unchecked, it can run rampant.

In the case of Bard, this led to a massive financial and repetitional backlash for Google and Alphabet. But there have been far more serious and damaging results off the back of machine learning and artificial intelligence.

In the US, artificial intelligence has been deployed for the automatic sentencing of drivers.

However, the original source data was shown to be heavily biased against non-white people who were stopped far more regularly than white people. It’s a sobering example of how terrible decisions can be made when data isn’t used and sorted properly.

With this in mind, there are some clear and present issues surrounding machine learning and artificial intelligence. Big tech organisations have a duty to ensure that the deep learning systems they develop and implement are not morally, societally or politically swayed in any way.

This is naturally a much larger and far-reaching matter, than the one of how ChatGPT will implement into teams, or change the way we search online.

Is the future bright or bleak?

Will AI take over the world? This is a question we’ve all been asking ourselves for longer than we care to remember.

The origins of artificial intelligence can be traced back to ancient philosophers. However, its origins as we know them date back to the fifties. So we should be under no illusion that this is a modern phenomenon.

We should also not kid ourselves, the machine learning and artificial intelligence haven’t already found their way into our day-to-day lives. Predictive text, Google Translate and even Uber all use machine learning and artificial intelligence and have been commonplace for some time now.

Just like the space race of the sixties boosted consumer technology, we’re now in a time of rapid and exciting technological development. Breakthroughs in machine learning are fuelling increasingly more powerful and capable artificial intelligence systems. Recent developments in quantum computing could bring forth even more powerful intelligent machines in the future.

In summary

The developments over the last 12 months have greatly advanced the cause of artificial intelligence and machine learning as a tool.

We’ve seen leaps and bounds, successes and failures. We’ve seen public outcry against ChatGPT managing to pass MBA examinations and we’ve seen creatives malign the use of Stable Diffusion. But these tools aren’t quite the existential threat that they are made out to be.

We’re optimistic that the future of artificial intelligence isn’t some Asomovian dystopia. Rather one where explicitly programmed machine learning models can be used in harmony with human-made thinking power.

Talk to us about your next steps, book a free strategy call at [email protected] or 0117 905 1177