Microsoft: ‘ever present’ personal AI assistants are coming

Artificial Intelligence (AI) assistants with “really good long-term memory” are about a year away, according to Microsoft’s head of AI, Mustafa Suleyman.

Products which can recall conversations, projects and problems will encourage users to invest more time and share more of their personal history with them, he said in an exclusive interview with the BBC.

“I think we’re moving to a fundamentally new age where there will be ever present, persistent, very capable co-pilot companions in your everyday life,” he added.

Critics have voiced strong concerns around this level of integration, including data security, privacy, the possibility of AI tools giving bad advice or wrong information, or displaying inbuilt bias towards the person they are supposed to be helping.

But AI supporters argue that in order to be truly useful, these tools have to be deeply embedded into our lives: that they can only be really helpful if they know the history and context behind what they are being tasked to do.

For example, an AI diary manager can only organise your diary if it can access that diary, edit it, and retain information about your activities.

Mr Suleyman argued that many people’s privacy expectations have changed over time.

He said that devices such as TVs, laptops, phones, in-car cameras and earbuds are already “recording continuously everywhere” in ordinary environments, and gave a further example of an iPhone feature called Live View in which video and audio is recorded at the same time as a photo is taken.

“Most people love that feature,” he said.

“Some people turn it off but that’s a very distinct shift in the default expectation of what a photo is.”

He added that the benefits of this kind of tech, whether people felt they could control their use of it, and whether they trusted the provider of it, were important factors in deciding whether to embrace it.

Microsoft has invested billions of dollars in OpenAI, the creator of ChatGPT, and has emerged as a market leader as the tech giants jostle in the race to develop and control the powerful and rapidly evolving technology.

But some research suggests people are not consistently using it. A poll published by the Reuters Institute in August found that 29% of people in the UK that it spoke to had used ChatGPT, but only 2% used it every day.

Mr Suleyman conceded that perhaps consumer AI tools would never be as globally popular as the smartphone.

“Maybe this is different to the smartphone,” he said.

“Nearly 90% of the planet has a smartphone. Maybe that will be different. Maybe 50% will reject [AI tools}.”

But he added that so far, AI had been the fastest growing and adopted technology in history, despite its potential risks.

He strongly rejected the idea, posed by many industry watchers including Jim Covello, head of stock research at Goldman Sachs, that AI might turn out to be a bubble, like some tech trends before it.

He told me about a woman he met who said she had set up her business using guidance and motivation from a chatbot he developed called Pi.

“We are clearly producing personalised, interactive knowledge at your fingertips at zero marginal cost,” he said.

“The idea that this could be a bubble is utterly beyond me.”

Microsoft has today unveiled a range of new additions to the tech giant’s AI assistant range CoPilot, including a voice function, a daily news digest and a slower chatbot for more difficult or in-depth questions called Think Deeper.

It also includes CoPilot Vision, a tool which will sit within its Edge web browser and, when activated, observe web pages and “assist” with online activity.

The firm says Vision will not record or store data, has to be switched-on manually and will close at the end of each browser session.

Microsoft says it has chosen to limit which sites it will work with and there is as yet no release date for it.

In the summer the tech giant paused the release of an AI tool called Recall, which takes screenshots every few seconds in order to help users find things they were looking at or working on previously, following a backlash from privacy campaigners and enquiries from the UK’s data watchdog about it.

It will be re-launched in November with additional security measures.