Kathy Gibson reports from the Microsoft AI Tour – Artificial intelligence (AI) is not some magic wand that will take over the world of work – and it’s certainly not smarter than people.
That is word from Scott Hanselmann, vice-president: developer community at Microsoft, who points out that AI is a tool that can be used to empower people, create jobs, make tedious jobs go away, and help more exciting and creative jobs to come about.
All new technology has always been hailed with excitement and a degree of trepidation, he adds.
“We need to push back on the idea that AI is smarter than us: it is artificial intelligence – not real intelligence or real creativity,” he says. “You need to push back on the people who say otherwise.”
What it is, he adds, is the next development in the ongoing journey of technology innovation.
“We went from client/server to the Internet to mobile and cloud – and now we have AI.
“What we have to do is ask ourselves how it will make our lives better. Because the whole point of new technology is not to make the stock price go up, but to make peoples’ lives better.
“If we are not doing that, we are doing something wrong.”
AI is only as good as the data it accesses, Hanselmann adds. “It doesn’t speak any language: it speaks numbers. So vague prompts give you vague answers.”
He points out that AI can produce useful results using the right data sets – and that these language models can be relatively lightweight, not necessarily the large language models (LLMs) that run on the world’s largest data centres.
“I always tell people to use the smallest and cheapest model you can to do the work.
“You can use CoPilot or open source, but think about what model you need to get the work done in the most cost-effective way.”
Pelonomi Moilloa, CEO and co-founder of Lelapa AI, has developed a tool that helps contact centre agents engage with customers in their home language.
Most AI – indeed most technology – has a built-in bias, she says. This bias is operating in English.
“Most models have been developed in a world that is not always aligned to our reality – and often they are being trained on data that is irrelevant to us.”
The team at Lelapa AI realised they could build smaller language models using smaller sets of relevant data.
Building on its experience in machine learning, the company built its AI model for contact centre agents.
“We believe people should be able to speak the language of their choice,” Moiloa says. “In South Africa, only one person in 10 speaks English at home, so a lot of the people who call into a contact centre are misunderstood.”
Lelapa’s API VulaVula takes what the customers asks in their own language and translates this to English for the contact centre agent. The response is then typed in English, whereupon it is translated into the caller’s language that a simulated voice reads back to the caller.
The system is built on Microsoft Azure, on the Azure Kubernetes service. With Microsoft Trustworthy AI safety, security, and privacy are built-in.
“I don’t think people realise how capable models can be even if they are smaller than those in the cloud,” says Hanselmann.