Former Google Designer reveals what’s behind AI models like Gemini

• Google has launched Gemini 2.0, taking it one step closer to creating a universal personal assistant.
• A former Gemini conversational designer talked about chatbot design best practices.
• He said Google’s AI products and its search engine run into problems of self-cannibalization.

Google launched its Gemini 2.0 model this week, promising more “agentic” AI to bring people closer to a version of a universal personal assistant.

When Google pushed out Gemini 1.0 last December, it was trying to compete with OpenAI’s ChatGPT. Gemini quickly changed how users experienced Google itself, from providing an overview of search engine results to the product NotebookLM, which can convert written notes into a spoken podcast. Its 2.0 version has features like “Deep Research”, where Gemini can scour the web for information and prepare reports.

As AI assistants become increasingly human-like in their delivery, the engineers and designers who build them must tackle the issues of responsibility and tone. For example, some AI chatbots may refuse to provide answers on potentially sensitive topics.

Business Insider spoke with Kento Morita, a former Google Gemini conversation designer and Japanese-American actor and comedian.

Morita has previously worked on designing conversation flows for Amazon Alexa and Google Gemini, specifically focusing on building a Japanese persona for AI. He provided insight into how AI chatbot designers think about delivering information to users effectively and the challenge Google faces in balancing its search engine and AI products.

The following was edited for length and clarity.

Business Insider: How are “tones” designed for sensitive topics for AI?

Kento Morita: Every time we get a question that might be sensitive, it goes through a sort of checklist like: is this political in nature? Is this sexual in nature? Does this generate something that is counterfactual and when? When the answer is yes, it goes through a process to ensure that all of these companies eventually have their logo next to the answer they give. Kinda like Warren Buffett’s rule of thumbwe should be happy to see that on the front page of The New York Times or the Washington Post the next day, and we should be proud of it.

The biggest question we need to answer is: Is it productive for their bottom line to associate Google or ChatGPT or anyone with this answer?

If it’s not, we do what’s called punting. We’ll just give one: sorry, I can’t help with that kind of answer now. It’s a balancing act. Some topics we don’t even want to touch with a 10-foot pole, but there are some things that we want to answer, like, for example, like election night coverage — everybody’s going to wonder what’s going on.

We want to make sure that answering more questions allows more people to stay on our website. There is always a tension in these companies to want to answer as many questions as we can, which any of these LLMs can, but it also has to be balanced with, will this create more negative press or will this provide answers that are potentially dangerous? A lot of talking to the legal team, talking to the marketing team, talking to sales. It’s an ongoing conversation all the time about how we want to approach it.

It is always a question of what to prioritize.

It is also a problem of cannibalizing a market.

One of Google’s biggest products is search. When you deliver Gemini, what does that mean for the search business? It is an ongoing existential question.

For companies like Google, companies like Perplexity AI can actually have an advantage here, I would say, because they’re in the business of creating one product and doing one product really well. They don’t actually run into self-cannibalization problems. I think really interesting things and really bold things are happening from companies that are not connected to a big conglomerate. I think that is only natural.

Google moved Gemini under the DeepMind organization. I really don’t know why they did this, but as a (former) employee and also someone who has followed Google for a long time, it is interesting that they are bringing many of the AI ​​companies under one organization, especially in light of the antitrust case going on right now around Google, and the conversation they’re having with the Department of Justice about whether or not Google should be broken up. At least if they split it up, I think they’ll have a conversation about how it would make sense to split it up. And having Gemini be part of an AI organization versus a search organization, I think makes sense.

We’ve been used to using Google search with ads at the top. Now it is The twins. It’s not the most factually up-to-date result, but it’s a shift.

The Google Search team is full of talented engineers. Their North Star goal is to provide search results that are relevant and accurate, and that has been their goal all along. And now you enter ads. Now enter Google Shopping results. Then you bring in Gemini. All these other factors in the organization intervene Google.com website design.

I wouldn’t be surprised if many of the engineers and people who have been working with Google search for the longest time are very frustrated. That said, I also wouldn’t be surprised if they welcome the idea of ​​breaking up the company to allow them to focus on what they’re excited about, which is delivering great search results.

Can you tell me about the story of add footnotes to chatbots and whether it was a conscious decision? How have hallucinations changed how chatbots react now?

Even with Google Assistant and Amazon Alexa, when you ask it a factual question, it used to immediately say, according to Wikipedia, blah blah blah blah, or according to XYZ, blah blah blah blah. At the time, it was quite difficult to convince people that it is a good idea. And the reason is that from a conversational point of view, when you ask someone hello, e.g. when was XYZ invented? You don’t really want to hear XYZ was invented, according to Wikipedia, in 1947. You just want to hear the answer. Getting to the answer quickly is considered a virtue in design. Google spent so much time and effort to make the time to display the search results as short as possible, so it has been in Google’s DNA to get the answer to the customer as quickly as possible.

We had to advocate for footnotes. What really convinced them was this idea that the moment you attribute to a site, you’re going to shirk responsibility for the accuracy of that information to another site.

So when I say, according to Wikipedia XYZ, I am no longer responsible for whether what I say is correct or not. I could just leave that responsibility to Wikipedia. And when people started asking sensitive questions about anti-Semitism or like conspiracy theories and what have you, being able to say, according to XYZ, that seems to be the case, it allows us to distance ourselves from that statement, which is very , very useful when we talk about Google’s brand image.

When you have something labeled Google Assistant and say that’s what happened, you can’t help but associate Google with what you’re talking about. So that kind of distancing language allows us to take less responsibility for the information that is presented. So I think that ethos has stuck, and those kinds of arguments have been really helpful in convincing people in those companies to cite our sources. Like Perplexity AI, because it’s so explicit in footnotes to everything, they actually have more freedom to talk about more controversial topics.

They don’t need to edit anything, which is really a huge advantage when it comes to controversial and sensitive topics.

Clarity is something they talk about a lot in the LLM space. LLMs, for many people, feel like a black box, like you type some text and it spits out text. But ultimately it is a prediction engine. Adding guardrails, editorializing, to content design around this black box that is a prediction engine has been very, very important, especially around sensitive information.

When Google Gemini and other artificial intelligence cites sources, is it still a prediction machine?

There is something called RAG (retrieval augmented generation). I think what they are doing is indexing sources like AP News and Reuters higher to skew those sources and the information in those sources higher. As LLM pulls more information from them, there’s an attribution mechanism in the background that allows them to say, “We’re using RAG to call Reuters or AP News to get their information.” I don’t think it’s predictable. It’s much more hard-coded.

For some topics, such as abortion, AI chatbots take on a caring tone, like asking, “Do you have any concerns?” It is a significant change of tone.

It is one of the biggest things that I feel very proud to be involved in. While we were developing Google Assistant, whatever words about suicide or self-harm came up, we went around talking to mental health professionals and people who offer these services and to ask them if we could give users a number for this hotline , no. 1, would that be helpful? No. 2, what is the best language to do it? We were very careful to talk to all of these resources.

For my part, I spoke with Japanese resources and Japanese hotline operators and we translated these messages. It took a lot of time, but we tried to make sure that every user, even users who were thinking about self-harm, gets the best information they can.

When it comes to abortion, it fits into the same strategy framework, content strategy: how do we ensure that people seeking abortion, how do we ensure that they get the information in a way that is safe and ultimately helps them to live the life they want? When I was at Google, we were able to fulfill our mission statement, which is to collect the world’s information and make it as useful and accessible as possible for everyone.

Ultimately, the democratization of these engines will happen. Every company will have a pretty decent LLM eventually in 5-10 years. The difference in whether I want to go to X or ChatGPT or Google or Alexa or whatever, the difference is going to be in the packaging.

The more these tech companies start treating people like people and make robots talk human, I think they’re going to be the most successful in the long run.