A disturbance in the force for ChatGPT
Big problems for applications that don't take into account the UX of AI
Chatbots are so 2022
ChatGPT is in trouble. And it’s not alone in that.
The AI UX revolution is in full swing, and you can’t bring a 2022 knife to a 2024 laser gunfight. I am not just talking about UIs like mind map generators like Mapify (although it is pretty cool and useful for a visual thinker).
Text-based interactions like chatbots are limited and interaction patterns like accordion editing and apple picking (as described by Jakob Nielsen in this article from September 2023, seemingly a decade ago) can only get you so far.
Voice-based interactions are more portable and require almost no visual UI but suffer from some of the same problems when it comes to extensive collaborations that accumulate content over time and require scrutiny of the outputs: they are very limited when it comes to information management.
Even today, if you want to search through your past chat sessions with Generative AI, there’s no easy tool to do that within ChatGPT or Claude (although there are some browser extensions like Echoes that promise to help you do just that).
Meanwhile, OpenAI's response with their Scarlett Johansson-esque voice agent that can convey a simulation of emotion, feels a bit like bringing a charming but ultimately unhelpful (and perhaps harmfully addictive) robot to a productivity fight.
Which doesn’t seem quite enough when you see what other competitors are bringing to the market.
The New Wave of AI Interfaces
Check out Anthropic’s Claude Artifacts and its “show and tell” approach to interactions, telling you what it did on one side of the page, and generating the code to show it to you in action on the other side. A few awesome use cases for this: quick prototyping of forms & pages, and interactive charts (flowcharts, org charts, mind maps, etc).
Ever wished you could chat with your messy collection of notes, links, and files? Google’s NotebookLM makes it possible. It's like that magical kitchen drawer where you can actually find what you need, or the Downloads folder that finally makes sense of all those whitepapers you've been "meaning to read." Not only that: they recently added a feature to let you listen to a conversation about your sources as if it were a podcast.
With a slightly different approach, Sana.ai is as a clever mix between a note-taking app and NotebookLM. It records, translates, and summarizes your conversations, then lets you augment them with templates, external files, and 3rd party app integrations.
The Microsoft Copilot Invasion
But the real game-changer? Microsoft's Copilot. It's not just coming for everyone's lunch; it's eyeing their pocket money and Pokemon cards too. Here's the lowdown on the recent deluge of Copilot integrations recently announced:
Copilot’s notetaking tool is natively embedded in Teams and was released a while ago, so people have already been experiencing it to generate meeting summaries, action items, lists of pending questions, etc (bye-bye Otter, Fireflies, Fathom, etc.)
Copilot Pages, an AI-powered collaborative canvas, allows several people to collaborate -with each other and with AI- on the same document.
Python for Excel brings data analysis to the masses, allowing you to surface insights and perform data manipulations without affecting the original data. Complex BI dashboards? More like "bye-bye" dashboards.
Copilot for Outlook, OneDrive, and Word: Manage your inbox, find files and content in your cloud-backed mess, and write directly in Word with AI assistance. No more ChatGPT copy-paste dance.
Narrative Builder for PowerPoint generates professional, brand-aligned presentations with outline, content, and layout suggestions. Gamma.app might want to watch its back.
And keep an eye on Copilot Agents, allowing you to build BizChat chatbots (similar to Custom GPTs) connected to Sharepoint as its source of data, which can be invoked from MS Teams when you have a question for them, or when you need them to do something for you.
The Elephant(s) in the Room
Of course, with great power comes... a whole lot of concerns:
Privacy: Will employees feel comfortable with Copilot eavesdropping on everything they write and say? Or will we see a rise in "off-the-record" conversations due to the loss of psychological safety?
Autonomy: Will these AI agents be genuinely useful, or just glorified paperclips?
Accuracy: Can we really trust the AI's suggestions, or are we headed for a "computer says no" scenario?
Governance: Empowering employees and democratizing access to information and tools is great. But who's keeping tabs on all these new tools to ensure consistency, coherence, efficiency, and proper use?
Security: Can we trust that Copilot will fully respect the level of access of each respective document and their information, or will it leak your confidential documents to whoever asks?
Intellectual Property: On whose data has this been (and continues being) trained? Do they know and approve of that use? Do you, in the case of your data?
The Road Ahead
While Microsoft seems to be gunning for world domination, it's not all smooth sailing. Copilot's jack-of-all-trades approach might leave room for specialized models in critical areas, an opening that other competitors might leverage.
And let's face it, if SharePoint has taught us anything, it's that "can do" doesn't always mean "easy to do." Users will need a lot of help getting this set up properly, securely, and ethically. That probably creates an opportunity for companies willing to act as implementation consultants, installing not only the tools but also the mindset of collaboration with AI.
OpenAI’s recent, very smart (for some things) o1 model may represent a temporary admission of defeat: maybe they will take the BYOUI (Bring Your Own User Interface) to AI approach, and focus more on the business of selling compute tokens to companies that focus on figuring out how these pesky humans will interface with the LLMs.
As we navigate this new frontier of human-AI interaction, one thing's for sure: we're witnessing an unprecedented pace of advancement coupled with an equally unprecedented lack of... well, precedent. It's like we've invented the equivalent to the Star Wars lightsaber, but we're still designing an ergonomic pommel that doesn’t fall apart, cutting your hand off, or figuring out why it sometimes cuts through blast doors and other times gets stuck on plot armor.
Welcome to the AI User Interface revolution. May the UX be with you.
Think you are spot on regarding most of the elephants (lol) here.
But regarding 2-3, don't you think that embedding it into Copilot will help with these to? E.g. spinning up Copilot Agents based on your own data would potentially increase accuracy and usefulness?