TEXTILES.ORG
Artificial Intelligence (AI) was one of the hot topics at the Advanced Textiles Association’s Expo this year. In his show floor presentation, “Unlocking AI: a small business guide to harnessing the power of artificial intelligence,” Adam Penner, AP Consulting and Technical Services, unpacked the Hollywood hype to consider how AI can be used effectively by small businesses —and he identified risks, as well.
Systems such as ChatGPT, Google Gemini and Meta LLaMA are trained using massive data sets; GPT-4 has 100 billion neurons and 100 trillion synapses, for example. Among its benefits are the ability to process large amounts of information, which means it’s ideally suited to tasks, such as reading a large technical manual and reducing it to something more manageable or pulling out just one aspect from it.
Writing text for letters, web or social media content was a popular discussion point with the audience of largely small-to-medium enterprises (SME); however, the consensus was that it is useful as a tool to support research and writing rather than a full replacement for human endeavour and experience. The ability to draw on vast quantities of data means that information can be gathered quickly and from across many sources in a way that an individual human would struggle to match, particularly when reaching across highly technical sources.
Identifying risks
The risks and negatives are also apparent, as Penner pointed to four distinct categories. The first is that it’s time-consuming and energy-intensive to train. People need to input the information, which means gathering and sorting data, then testing what happens when a person tries to retrieve information.
A human with years of experience in a field might pick up on gaps or missing data, while a machine will need to learn that in itself. In some instances, a reasonable guess can be made based on like-information, but in other scenarios it can literally mean the difference between life and death. Relying on AI to know the difference each time creates risk.
For companies cautiously guarding their carbon footprint, the power needed is not good news. Microsoft has signed a 20-year power supply deal with energy provider Constellation that will see the reopening of the Three Mile Island nuclear plant in Pennsylvania, which was the location of the most serious nuclear accident in American history. It appears that the industry is indicating its lack of confidence in renewables or batteries to provide sufficient energy to meet future needs.
The second issue is based on ethical concerns. Data has to be put in by a human, so their bias will follow that information and flow out again. The third concern has to do with privacy and data ownership. This is not new, but it is telling that progress has been slow. When considering use-case scenarios in medical and healthcare among especially vulnerable and elderly groups, this is a serious worry.
Finally, there are “hallucinations.” This is an interesting use of the term, but essentially it means that the AI takes data and proceeds to imagine things, or it hallucinates. This comes as something of a surprise given the questions raised over the years about its potential for creativity and imagination. The phenomenon can be more concerning when considering the potential results of an hallucination while processing a training or safety manual.
Dr. Marie O’Mahony is an industry consultant, author and academic. Based in London, she is a regular contributor to Textile Technology Source.