AI comes for the Office
AI comes for the Office
Google and Microsoft are adding AI chatbots to their productivity suites, which could tell us a lot about where artificial intelligence is going next.
This week was nothing short of an artificial intelligence extravaganza. We saw the unveiling of GPT-4, OpenAI’s newest version of the AI model that powers the ever-so-popular ChatGPT application. We saw examples of how powerful the new tool is, what it means for contextual AI conversations moving forward, and the challenges it’ll face as it hits the mainstream as a direct-to-consumer product.
Of course, we also heard from both Google and Microsoft, who made two of the biggest announcements - arguably - in their respective histories.
It’s in these announcements that we get the biggest story of the week: AI is coming for the office. One of the most popular use cases for ChatGPT has been office work, where people are generating emails, job descriptions, essays, and more using OpenAI’s technology and touching it up afterward. There are even sites publishing full-blown articles generated by AI.
Given the fact both Google and Microsoft have productivity suites that are used by hundreds of millions of people, it was safe to assume that eventually, AI chatbots would directly integrate with them. And that’s the news we got this week.
Google AI assistant for Workspace
Google was first to unveil its AI assistant for Workspace, which oddly doesn’t come with its own branding. (I checked to see if the assistant was powered by Bard, but nowhere in Google’s press release is that name mentioned.)
The new assistant seems chock-full of neat features that can enhance any workflow, whether you’re trying to get a head start on a lengthy document or need to summarize meeting notes. Google supplied a list of the things it’ll be able to do once it’s fully rolled out, such as:
draft, reply, summarize, and prioritize your Gmail
brainstorm, proofread, write, and rewrite in Docs
bring your creative vision to life with auto-generated images, audio, and video in Slides
go from raw data to insights and analysis via auto completion, formula generation, and contextual categorization in Sheets
generate new backgrounds and capture notes in Meet
enable workflows for getting things done in Chat
The most notable new feature is AI text generation in Gmail and Docs, which will let you kickstart new documents or give you an idea of what to say in an email. It can generate a starter for what you’d like to communicate, or write the entire thing for you. This is very reminiscent of ChatGPT, except powered by Google’s intelligence systems with its own ethics codes and data pool. We haven’t seen any examples yet of how good Google is at AI text generation, but I suspect it’ll at least be on par with GPT.
Using the assistant for sorting data in Sheets, summarizing meeting notes, and automatically generating new Slides also sounds promising, so I’ll be curious to see how well each feature works once they roll out.
Microsoft Copilot
Then there’s Microsoft, who announced a similar set of tools for its Microsoft 365 suite of apps named “Copilot.” Unlike Google’s assistant, Copilot will be directly powered by OpenAI’s GPT-4, and on the surface, it looks like it could truly be revolutionary for Office users.
Copilot will live on its own across each Office application including Word, Excel, PowerPoint, Outlook, and more, springing to life whenever you need a touch of AI in whatever workflow you find yourself in. It’s sort of a modern take on Clippy, a productivity assistant Microsoft introduced in 1996 that grew a cult following after it began fading away in 2001. Whether Copilot can achieve that status is unclear, but it seems to be a lot more helpful in your everyday tasks.
Microsoft says Copilot will offer a wide range of functionality, from drafting documents in Word to generating 10-slide PowerPoints based off that same document. You can manipulate data in Excel (like generate tables, compare metrics, model outcomes based on variables, and more), catch up on your inbox in Outlook with a few commands, and even summarize meetings held in Teams.
For business users, Copilot turns into Business Chat which can help you catch up on new hires, check in on who’s back from vacation, prepare materials for meetings with staff, and more.
These features, at first glance, are massive additions to the smorgasbord of features already available across Microsoft 365. They have the potential to greatly enhance productivity, efficiency, and ease of use across each application Copilot is available in, and it’s pretty enticing to think about all the different ways you could use it to get work done. I imagine I’ll use it most in Excel since - surprise, surprise - I suck at Excel.
OpenAI GPT-4
As I mentioned before, Copilot is primarily powered by OpenAI’s GPT-4 model, which was also announced this week. It’s an improved version of GPT-3.5, which hit the mainstream as part of ChatGPT in December, taking the world by storm and turning the consumer technology industry on its head. Microsoft was quick to implement the tech into Bing and bring the search engine back to life, while Google had to call a “code red” internally and rush to introduce its Bard competitor that has yet to see the light of day (publicly, at least).
Its improvements seem pretty iterative - it’s “more creative and collaborative” than before, it supports both text and photo input to generate responses, and it can emulate your writing style a bit better - but it’s the foundational work OpenAI has done that’s most important.
This includes the rigorous testing phases it went through before being announced. OpenAI said that GPT-4 was in testing for six months, resulting in responses that were accurate 40 percent more of the time compared to GPT-3.5. It was also 82 percent less likely to respond to requests for content that went against its guidelines.
It’s also impressively smart. OpenAI says it had the new model take the Uniform Bar Exam, LSAT, SAT Math, and SAT Evidence-Based Reading & Writing exams, and it managed to score in the 88 percentile and above across each test. That’s extremely impressive for an AI model, especially one that’s as mainstream as it is.
GPT-4 is also “multi-modal,” which basically means it can generate responses via other mediums besides text. In the future, this could mean that videos and complex photo edits will be generated using ChatGPT, but for now, it can only understand combined text and photo input and generate text-based responses.
Between Google’s AI assistant and Microsoft’s Copilot, it’ll be interesting to see which offers a more seamless, intelligent experience. Neither will be perfect out the gate, as one might assume. Google reminds us that “sometimes the AI gets things wrong, sometimes it delights you with something offbeat, and oftentimes it requires guidance.” Meanwhile, GPT-4 has secretly been powering Microsoft’s new version of Bing since it launched last month, and many people have manipulated it to a point where it says some pretty crazy things. OpenAI is also not hiding the fact that GPT-4 can generate hateful, violent text at times and completely mislead people with factual errors. But like versions of GPT before it, it’s all a work in progress, and both OpenAI and Microsoft are well aware of the risks associated with testing it out in the open.
It’s still the early days of AI chatbots taking over the everyday products and services we use, and this foray into the world of productivity suites will serve as a meaningful - if not historical - testing phase for what the world actually wants from these incomprehensibly powerful super computers. Maybe it’s generating a good email to send to your boss to ask for more vacation time in Gmail, maybe it’s sorting through piles of data in Excel, or maybe it’s none of the above. The market will dictate where AI goes next, which makes these product rollouts all the more exciting.
Footnote
I’m sure you’re all wondering when you might be able to use Google’s AI assistant or Microsoft Copilot. Unfortunately, I don’t have much to tell you. Google seems to be thinking about releasing a few features during the remaining nine months of 2023, while Microsoft is testing Copilot with 20 different users and plans to expand in the future. If I were you, I’d bet on both features to roll out by the end of the year, if either company’s history in this department is any indication.
In the meantime, GPT-4 is available for anyone to use, as long as you can beat the demand.