Suggestions
Jonathan Melhuish
Senior Treasury Manager at The Home Depot
Jonathan Melhuish is a skilled professional who excels in utilizing qualitative research techniques, lightweight prototyping, and his extensive knowledge of IT and software development, notably Artificial Intelligence, to assist clients in creating successful and exceptional products and services. With a focus on optimizing development resources and enhancing customer experiences, he delves deep into understanding customer problems and the business landscape to derive innovative and straightforward solutions. Jonathan specializes in aiding businesses in gathering data essential for implementing impactful AI-driven customer experiences. Available for freelance assignments, consultations, and training, he operates both remotely and on-site throughout the EU, with a keen interest in sustainability and climate change initiatives. Jonathan obtained a BSc in Artificial Intelligence and Computer Science with a 2:1 degree from the University of Birmingham. He has held significant roles such as Chief Executive Officer at Makuma, and previously served as a Service Designer at Verkla and as a UX, AI, and Customer Experience Consultant while working as a freelancer.
Highlights
It seems everyone's making fun of Google this week for its heavy-handed attempts to enforce ethnic diversity on its new Gemini image generation feature. But it's probably more helpful to ask: what can we learn from their blunder?
The success of AI systems depends heavily on the quality and diversity of the training data you provide to it. Inevitably what Google found is that if you train a neural network on images - e.g. stock photo libraries - that feature a lot of attractive, slim white people, that's exactly what you get as an output.
In this case, whether that's a problem or not is up for debate. The challenge is that in many other cases where we apply AI, biases in the data won't be nearly as obvious in the output. It's much harder to understand why, say, a decision support tool favours options that the CEO would approve of - perhaps because that tends to be the way that things have been done previously? Without a deeper understanding of the data, it's difficult to get an AI system to do anything other than repeat the past.
And, as the Google example shows, don't expect to slap a "sticking plaster solution" on it afterwards. The results might be embarrassing... or worse.
Want to check your training data is unbiased? Comment here or DM me.
Want more like this? Sign up to my newsletter: https://aiforprofit.com <<
When will we achieve Artificial General Intelligence (AGI)? Should you even care?
In some ways, it doesn't matter, because "human level intelligence" is a somewhat arbitrary and vague yardstick. Computers might never match every form of "intelligence" we have, simply because it's not that useful. Humans might be better at knowing when somebody is standing behind you. A robot can just have a camera in the back of its head.
The real value of AGI as a concept is giving us a sense of how fast AI is progressing across multiple domains. Computers have been better than all humans at chess since the 90s, have beat the average human at writing in the last couple of years, and will very soon be able to create videos rivalling a competent film crew.
Does that mean we achieved AGI? No. But it gives us a picture of how fast AI is progressing, and, in general, it's faster than we thought - probably because humans aren't great at estimating exponential curves. The above graph shows (with the x axis on a log scale) the predictions of experts about when AGI will be achieved. The punchline is: as we progress along the exponential curve of progress, their estimates keep getting shorter.
The reality is, none of this matters to your business. AI offers significant value in automating routine tasks today, and it will offer even more value in automating an even greater diversity of tasks tomorrow. Regardless of exactly how fast it moves, the time to act is now.
Image source: https://lnkd.in/dCtfCTud
