Revolutionizing Human-Computer Interaction: Microsoft’s AI Boss Unveils Copilot’s Emotional Support for Windows and Office

To go back to this article, navigate to My Profile, and then look for saved stories.

Microsoft's AI Leader Aims for Copilot to Offer 'Emotional Support' in Windows and Office Environments

Previously, Mustafa Suleyman played a pivotal role in a groundbreaking artificial intelligence movement.

As one of the founding members of DeepMind, a UK-based firm that became a part of Google in 2014, he pioneered a novel strategy enabling computers to solve extremely challenging tasks by integrating learning through trial and error with rewards and penalties. This method was showcased through DeepMind's creation of AlphaGo, a highly advanced program for playing Go, which outperformed the top global Go champion in 2016.

Currently, Suleyman is promoting a novel advancement in artificial intelligence.

As the chief executive officer of Microsoft AI, Suleyman leads initiatives to incorporate the artificial intelligence technology behind ChatGPT into applications, including the Windows operating system, which is utilized by the majority of personal computers globally.

Today, Microsoft unveiled an update to its AI assistant, Copilot, introducing a voice that mimics human speech, enhanced screen-reading capabilities, and improved logic functions.

Suleyman mentions that the strategy aims to reignite users' passion for personal computers. He discussed this with Will Knight, a senior writer at WIRED, during a conversation from Redmond, Washington, via Microsoft Teams. The dialogue underwent minor edits.

Will Knight: What's the fresh perspective for Copilot?

Mustafa Suleyman: We're currently experiencing an incredible moment of change. Artificial intelligence partners can now visualize what we visualize, listen to what we listen to, and converse in the same language we use to interact with each other.

This innovative design material focuses on durability, connectivity, and emotional engagement. I'm in the process of creating experiences centered on enduring, continuous interaction with a partner.

You transitioned to Microsoft from your previous role at Inflection AI, where you concentrated on developing AI systems characterized by their helpfulness and understanding nature. It appears you've imported this ethos into your position with your current company.

From my early beliefs, even before my time at DeepMind, I have been convinced of the capability of AI to offer assistance. In fact, one of my initial projects at the age of 19 involved launching a phone-based counseling service, focusing on providing emotional support.

This moment in technology is truly remarkable. Experiencing the interaction with these innovations for an extended duration—this partner that genuinely learns about you—is something special. It offers guidance, motivation, support, and education. I believe that it will no longer seem like interacting with a computer.

What's the concept behind Copilot Vision, the experimental feature accessible to Pro users for testing?

The vision feature allows you to inquire, "What is that object there [on your display]?" Or, "Hold on, what is this? What's your opinion on it? Do you find it interesting?"

Spending time in front of your computer is filled with numerous brief instances. Having an AI partner that can view what you're seeing and converse with you instantly about it is quite remarkable. This alters the path you navigate through your online experiences, as it removes the need to type out your queries or thoughts.

This seems reminiscent of Recall, the contentious and currently optional Windows function that captures users' screen activity.

With Copilot Vision, none of your data is stored; thus, when you end your session by closing the browser, everything is completely erased. However, I am contemplating whether and in what manner we might implement data retention in the future since there is a significant demand for such a feature. Imagine being able to recall, ‘What was that image I came across online recently? What meme was that?’ It's something we'll likely consider exploring eventually.

Currently, the Copilot Vision tool is in a transient phase. We will need to continually test and determine what is practical as we move forward.

How do the concerns around confidentiality escalate when individuals disclose personal data to Copilot?

The records of your interactions are securely maintained to meet the top-tier Microsoft Security standards. This is done to ensure that your conversation history is preserved, as it's clear that retaining these records is important to you.

Additionally, you're launching Think Deeper, a feature enabling Copilot to address more complex challenges. This is built upon OpenAI's o1 model, also known as Strawberry, correct?

It's akin to Strawberry, indeed. We've adjusted an OpenAI model to better suit our consumer-focused needs, ensuring it behaves in alignment with our AI companion motif.

How do they differ?

OpenAI primarily concentrates on solving complex scientific and mathematical problems. Our approach has been to direct its attention towards conducting comparative analyses and evaluating products and services from a consumer perspective, among other things.

When you encounter a challenging issue or need to meticulously think through a matter, it's beneficial to systematically compare options or conduct a comprehensive analysis.

Are individuals at Microsoft currently utilizing this latest iteration of Copilot?

Absolutely, it's become widely adopted. The entire organization had access to it just a few days back. Now, everyone is actively engaging with it and sharing a plethora of insights. Our channels for receiving feedback are completely overwhelmed. It's quite an enjoyable experience.

Individuals will likely recall Clippy, the final AI assistant from Microsoft for Windows. Do those in the know notice any similarities?

Funny story, I ran into Bill Gates recently, and he pointed out that we've been calling AI by the wrong name. According to him, it should be referred to as Clippy. My reaction was just, wow!

It really highlights the astonishing foresight individuals like Bill possess. They envision the future, not merely a couple of years down the line, but two decades ahead.

Do the latest additions represent progress toward the development of AI assistants capable of performing helpful tasks on a computer?

Certainly. The initial step involves AI interpreting the same data as you do – observing what you observe, listening to what you listen to, and absorbing the text that you read. The next stage consists of AI developing a lasting, continuous memory which fosters a collective comprehension over time. The final phase sees AI engaging with external entities by issuing commands and executing tasks – such as making purchases, scheduling appointments, and organizing plans. Currently, we are experimenting with these two capabilities in a research and development setting.

Hold on, do you possess an AI tool for Windows capable of making purchases on your behalf?

We're not there yet, but indeed, we've completed the cycle and executed transactions. The issue with this technology lies in its inconsistency; achieving functionality about half the time is possible, but pushing that success rate to 90 percent requires significant work. There have been impressive demonstrations where it autonomously completes a purchase, among other things. However, there have also been instances where it completely fails to perform as expected.

Can you provide more details about the car accident? Was a Lamborghini purchased using Bill’s credit card without permission?

It would be rather amusing if it charged Bill's credit card. However, as previously mentioned, we are methodically working through it. It's heavily under development in the research facilities. Progress is being made, but it's measured in months rather than years, in my opinion.

What do you see as the greatest obstacle in transforming the vision of AI you have described into actuality?

The key challenge lies in creating technology that garners trust, particularly because it will provide an experience that feels incredibly close and personal. It's essential to nail down the aspects of security and privacy, undoubtedly. However, the core issue is shaping the interaction in such a way that the digital assistant can define its limits, making it clear when it cannot participate in certain discussions.

Achieving that would lay the groundwork for a reliable experience. Once we have that trust established, we can delve into the more complex aspects, such as enabling it to make purchases for you, negotiate deals for you, sign contracts for you, or organize a multi-stop itinerary for a Saturday afternoon. You would feel confident enough to say, "I trust you, Copilot, you've got this under control." That's our ultimate goal.

Discover More…

Direct to your email: A selection of the most captivating and unusual tales from the archives of WIRED.

Elon Musk is considered a threat to national security.

Discussion: Meredith Whittaker Aims to Disprove Capitalist Principles

What is the solution to a dilemma such as Polestar?

Gathering Alert: Be part of The Major Dialogue happening on December 3rd in San Francisco.

Additional Insights from WIRED

Insights and Tutorials

© 2024 Condé Nast. All rights reserved. Purchases made through our site on products affiliated with our retail partners may generate revenue for WIRED. Reproduction, distribution, transmission, storage, or any form of usage of the content on this site is strictly prohibited without explicit written consent from Condé Nast. Ad Choices

Choose a global website