What does AI do, exactly?
What does AI really do? “Artificial Intelligence” conjures up images of a jolly plastic pal in the workplace, or, on the flip side, a controlling mind that will eventually decide the fate of humanity. But AI is an umbrella term for different tools and processes that replicate aspects of human intelligence. Many of those tools are already in common use. At last week’s AICPA & CIMA SME Conference on The Future of Technology in Finance we had some spirited discussions about the inevitable march of AI and how it has been absorbed into our normal lives.
The recommendations on your Netflix, Amazon or Spotify account are generated by machine learning. Your watching, buying or listening activity is processed through the machine learning model to show you what others with your exact profile have enjoyed. As your activity increases, choices loop back into the recommendation model and it learns to refine its responses. At work, your spam filters and antivirus and accounting software are learning, too. Strictly speaking all of the tools below have their roots in machine learning, whether supervised (pre-trained), unsupervised or deep learning models. However, the process is most obvious to the naked eye in marketing personalisation.
Are you using an app to transcribe your meeting notes? Have you asked Word or Google to read back what you just wrote? Have you asked Alexa, or Siri, or Cortana, to do something for you today? Your speech is being interpreted by AI. Every word spoken is tagged and compiled and rendered as text for transcription or to prompt a verbal response. Yes, the machines make mistakes – but they learn as the data set expands. Speech recognition is not without its challenges. Because AI needs a huge volume of data to learn from, native English voices of a certain pitch are easily interpreted. As the available number of audio recordings in the data set fall (think accented voices and less common languages) the accuracy drops markedly. This is an incredibly useful tool for accessibility and speeding up business processes, but it’s a long way from perfection.
After last week’s conference, I arrived back at my local station and went to pay for the car park. The machine showed me a picture of my car arriving that morning and calculated the charge. The barrier recognised my license plate and opened to let me out. This is AI in action, and we have come to accept it as the norm. Machine vision is under the hood of a swathe of industrial processes from product sorting to visual inspection for quality control, maintenance and safety, and gives eyes to autonomous vehicles of all types that need to recognise the world around them. AI vision technologies also give us widely-used image recognition. We’ve come a long way from mistaking cats for pandas on Google Image. Next time you try to identify a plant with an app, consider what technology you’re using.
When search engines took their first baby steps online, we learned how carefully we had to parse our requests to get any sense out of them. Those days are gone. We can throw a stream of consciousness into the search bar and come back with exactly what we were looking for. NLP can extract and classify content from text. It answers questions and generates text, images, graphics and code – think chatbots and customer service, as well as the now ubiquitous Generative Pre-trained Transformers, GPTs, that lie beneath ChatGPT, Bard, Dall-E, Midjourney and other common tools. NLP makes our interactions with AI feel human, but remember, these neural networks are only as good as their data, and they’re not thinking machines.
If ChatGPT is the generalist AI for every query, expert systems are the polar opposite. They focus on capturing human expertise to solve complex problems in specific domains. Healthcare is one area where expert systems have proven their worth, digging deep into vast quantities of very specific data that could not be processed by humans alone. Healthcare expert systems improve diagnosis for complex conditions, particularly at an early stage, improve clinical outcomes, and speed up drug discovery. In chemistry, complex molecular structures can be identified by DENDRAL, an expert AI system. In business, highly complex decision making around resource allocation, operational planning, and even fraud detection using expert systems is well established.
It’s very likely that some part of your life has been touched by aspects of Artificial Intelligence. How many of these examples have you encountered at home or at work in the last few days? We have a very powerful toolkit at our disposal, and the possibilities are endless.
Originally published by Kate Baucherel www.galiadigital.com. Kate is a speaker, author and consultant specialising in Web3 technologies including blockchain, cryptocurrency and AI.
