Discussions around future employability tend to highlight the unique qualities that we, as humans, value. While we might pride ourselves on our emotional intelligence, communication skills and creativity, it leaves a set of skills that would have our secondary school careers advisors directing us all off to retrain in nursing and the creative arts. And, quite honestly, if I have a tricky email to send, Chat GPT does a much better job at writing with immense tact than I do.<\/p>\n\n\n\n
Fortunately for us all, these simplifications of such a complex issue overlook some reassuring limitations built into the Transformers architecture, the technology that the latest and most impressive generation of AI is built on. <\/p>\n\n\n\n
These tools have learnt to be literate in the most basic sense<\/a>. They can predict the next, most logical, token that will please their human audience. The human audience can then connect that representation to something in the real world. There is nothing in the transformers architecture to help answer questions like \u2018Where am I right now?\u2019 or \u2018What is happening around me?\u2019\u00a0<\/p>\n\n\n\n In business these are often crucial questions. The architecture can\u2019t just be tweaked to add that as an upgrade. Unless someone has already built an alternative architecture in secret somewhere in Silicon Valley, we won\u2019t see a machine that combines Chat GPT with contextual understanding any time soon<\/a>.\u00a0<\/p>\n\n\n\n We, of course, have some wonderful examples of technologies that can actually answer questions like \u2018Where am I and what\u2019s going?\u2019 Your satnav, for one, has some idea where you are and of some hazards ahead. More traditional neural networks can look at images of construction sites and spot risk hazards before they become an accident. Machines can look at medical scans and see if cancer is or is not present. <\/p>\n\n\n\n But these machines are highly specialised. The same AI can\u2019t spot hazards around my home, or in a school. The machine that can spot bowel cancer can\u2019t be used to detect lung cancer. This lack of interaction between highly specialised algorithms means that, for now, AI still needs a human running the show. They must choose which machine to use, and whether to override the suggestions that the machine makes.<\/p>\n\n\n\n And that is the other crucial point. Many of the algorithms that are being embedded into our workplace have very poor understanding of their own capabilities<\/a>. They\u2019re like the teenager who thinks they\u2019re invincible because they haven\u2019t experienced failure and disappointment often enough yet.\u00a0<\/p>\n\n\n\n If you train a machine to recognise road signs, it will function very well at recognising clean, clear road signs. We would expect it to struggle more with \u2018edge\u2019 cases. Images of dirty, mud-splattered road signs taken at night during a storm, for example, trip up AI where humans succeed. But what if you show it something completely different, like images of foods?\u00a0<\/p>\n\n\n\n Unless it has also been taught that images of food are not road signs and need a completely different classification, the machine may well look at a hamburger and come to the conclusion that – of all the labels it can apply – it most clearly represents a stop sign. The machine might make that choice with great confidence – a circle and a line across the middle – it\u2019s obviously not a give way sign! So human oversight to be able to say, \u2018Silly machine, that\u2019s a hamburger!\u2019 is essential. <\/p>\n\n\n\n It does not mean the end of your career, unless you are in a very small and unfortunate category of professions. But it does mean that the most complex decisions you have to take today are soon going to become the norm. The ability to make consistent, adaptable, high quality decisions is vital to helping your career to flourish. <\/p>\n\n\n\n Fortunately for our careers, the world is unlikely to run out of problems to solve any time soon. <\/p>\n\n\n\n With complex chains of dependencies and huge volatility in world markets, it\u2019s not enough to evolve your intelligence to make more rational decisions (although that will always help – we are, by default, highly emotional decision makers). <\/p>\n\n\n\n To make great decisions, you need to know what you can\u2019t compute, and what the machines can\u2019t compute. There will be times when external insights from data can support you in decision making. But there will also be intermediaries to coordinate, errors to identify, and competing views on solutions to weigh up. <\/p>\n\n\n\n All machine intelligence requires compromise, and fortunately, that limitation leaves space for us, but only if we train ourselves to work in this new professional environment. At the Institute of Analytics, we work with professionals to support them in this journey.\u00a0<\/p>\n\n\n\n Dr Clare Walsh<\/a> is a leading academic in the world of\u00a0 data and AI, advising governments worldwide on ethical AI strategies. The IoA<\/a> is a global, not-for-profit professional body for analytics and data professionals. It promotes the ethical use of data-driven decision making and offers membership services to individuals and businesses, helping them stay at the cutting edge of analytics and AI technology.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":" Dr Clare Walsh, Director of Education at the Institute of Analytics (IoA), explores the practical implications of modern generative AI. <\/p>\n","protected":false},"author":480,"featured_media":22248,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"apple_news_api_created_at":"2024-12-06T14:45:41Z","apple_news_api_id":"be0b0203-cc58-48fb-8c59-bd8649d0ba20","apple_news_api_modified_at":"2024-12-06T14:45:41Z","apple_news_api_revision":"AAAAAAAAAAD\/\/\/\/\/\/\/\/\/\/w==","apple_news_api_share_url":"https:\/\/apple.news\/AvgsCA8xYSPuMWb2GSdC6IA","apple_news_cover_media_provider":"image","apple_news_coverimage":0,"apple_news_coverimage_caption":"","apple_news_cover_video_id":0,"apple_news_cover_video_url":"","apple_news_cover_embedwebvideo_url":"","apple_news_is_hidden":"","apple_news_is_paid":"","apple_news_is_preview":"","apple_news_is_sponsored":"","apple_news_maturity_rating":"","apple_news_metadata":"\"\"","apple_news_pullquote":"","apple_news_pullquote_position":"","apple_news_slug":"","apple_news_sections":[],"apple_news_suppress_video_url":false,"apple_news_use_image_component":false,"footnotes":""},"categories":[3],"tags":[],"topic":[614],"class_list":["post-22247","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-the-interface","topic-data-ai"],"acf":[],"apple_news_notices":[],"yoast_head":"\n
Where transformers have been revolutionary, it tends to be areas where humans had almost given up the job. Medical research, for example, is a terrifically expensive and failure-ridden process. But using a well-trained transformer to sift through millions of potential substances to identify candidates for human development and testing is making success a more familiar sensation for our medical researchers. But that kind of success can\u2019t be replicated everywhere.<\/p>\n\n\n\nJoining it all up<\/strong><\/h3>\n\n\n\n
AI: Confidently wrong<\/strong><\/h3>\n\n\n\n
What does this mean for the next 10 years of your career?<\/strong><\/h3>\n\n\n\n