Gemini Deep Research

In today’s episode, we'll explore the features and functionality of Google's Gemini Deep Research as well as ways you might use it in your work as a teacher. Visit AVID Open Access to learn more.

Transcript

Paul Beckermann 0:01 Welcome to Tech Talk for teachers. I'm your host, Paul Beckermann.

Transition Music with Rena's Children 0:05 Check it out. Check it out. Check it out. What's in the toolkit? Check it out.

Paul Beckermann 0:16 The topic of today's episode is Gemini, deep research. Deep research is a new option in Google's Gemini AI chat bot that's getting a lot of people excited, especially those who want high-quality answers to their queries and answers that are grounded in research rather than general web content.

In today's episode, I'm going to give you an introduction to what this tool is, what the process and product are like, and how you might use this in your work.

Transition Music with Rena's Children 0:44 Here's the tool for today.

Paul Beckermann 0:48 So first of all, what is Gemini deep research? To answer this question, it's helpful to think of deep research as a super-powered research assistant fueled by Google's Gemini AI. It's a feature built into the Gemini AI chat bot designed to go beyond simple searches.

To do this, it acts like a researcher and strives to actually understand complex information, synthesize it, and present it back to you in a way that's useful for in-depth analysis. Rather than just pulling up links, it's reading, comprehending, and drawing conclusions from vast amounts of data, much like a human researcher would do, but it's doing it much faster than humanly possible.

To help you get a better understanding of the experience of using Google's deep research, I'm going to walk you step-by-step through an interaction that I had with the platform. Step one: accessing deep research. To access this feature, you need to go to Google's AI chat bot, Gemini, and log in.

Once you're logged in, you should see the deep research option listed below the text input field in the chat bot. Click on deep research to launch it.

Step number two: identify source information. Before you initiate a research task, you have the option of specifying where deep research will look for answers. By default, it will search the web, accessing diverse sources, such as academic journals, scientific reports, reputable news articles, government databases, and even potentially synthesize information from multimedia sources.

However, you do have the option of customizing this by clicking the sources drop-down menu. You can also choose to have deep research look through your Google Drive, Gmail, and chat. If you select these services, you'll need to connect them to your Gemini account to provide it with access to that content. Another option you have is to upload files.

If you click the files button, which is next to the sources drop-down menu, you can upload up to 10 files using the free version of Gemini. Gemini accepts most file types, although there are some size and quantity restrictions when using the free version. While the paid version will give you considerably more capacity for uploaded content, the free version is a great place to start. This upload option makes the experience feel a bit like NotebookLM, where you can limit the scope of the research by asking the chat bot to reference specific studies or documents that you provide.

Step three: enter your prompt. This is a key step, and it's similar to writing any generative AI prompt. You want to assign the AI a role, describe clear tasks, identify an audience, specify details about final outputs, and add relevant limiters and context to improve the results. It can be helpful to describe the research parameters you would like to use as well.

Perhaps you want to require deep research to only reference peer-reviewed sources, or maybe you want the output to be in the form of a literature review or perhaps a pro/con table. The more detailed you are in your prompt, the clearer target the AI will have for its work, and the more satisfied you will likely be with the result.

To get started, I typed in: "Act as a K-12 educational research expert. Research best practices in using generative AI in K-12 schools. Focus on US schools during the past two years. Generate a list of top 10 most popular and effective strategies." Then I hit ENTER to initiate the AI.

Step four: review and edit the research plan. This is where deep research starts to feel different. Rather than just giving you a response, it shows you its research plan for you to review and revise as needed. This is a big step toward using AI as a true collaborator, rather than just an agent to do work for you.

It allows for more transparency and clarity about what the AI is doing and where the content is coming from. For my prompt, it outlined how it was going to research websites to get the best information. It showed me a seven-step process that included tasks such as searching for policy, pedagogy in US schools, studies and pilot programs, common models of professional training, best practices for ensuring academic integrity, compiling a list of top generative AI tools, and then synthesizing the collected data to categorize and prioritize strategies based on adoption rates and positive educational outcomes.

When I first reviewed that process, I was literally blown away. I felt like I was back in my master's level research class. It presented a really high-quality research plan, but that wasn't the end of the plan. Below that, the AI also outlined that it would analyze results and then create a report. There's even an edit button available.

If you want to modify any of this, simply click that button and explain what you'd like changed. The plan will be revised accordingly and leave you in the loop as part of the research process.

Step number five: start the research. When you click Start research, the AI gets to work executing its research plan. It opens a window on the right side of the screen where you can watch the progress. One major difference from regular AI chat bots that you will notice is that deep research is not as fast. In fact, it can feel quite slow when compared to a web search or even a typical AI chat bot query. Because you are asking it to conduct deep research using sophisticated reasoning and thought processes, it will take a while. You're essentially trading off a little extra time for better quality content.

For my example, it took about five minutes to generate a report. Considering the results I got, I felt like this was actually very fast.

If you don't have the time to wait, you can close the browser as it works and move on to other tasks. In fact, you can even shut down your computer and it'll keep working in the background of your account. You're logged in, so it knows to store the results for you when you return to your account.

I found it really interesting to watch the AI work. It posts updates on what it's doing as it works through the research plan. For instance, it reported that it was reviewing a specified collection of web links. It listed those links. It revealed that it was synthesizing its findings and determining gaps in the information based on its review and internal self-analysis. It then completed a very iterative process and moved on to what it deemed to be the next logical step in completing the task. And the cool thing is, you can watch it work as it goes through the process.

I feel like this takes some of the mystery out of AI, and it gave me a lot more confidence in the results.

Step six: view the report. When deep research is finished with your task, it will generate a nicely organized and formatted report. There is a convenient drop-down menu option at the top titled "Contents." By clicking this, you will see an outline of the report. Click on a section to jump to that spot, or simply scroll down and review the report from top to bottom. If you click on a section, it will bring you to an editable view so you can work with the generated content and refine it as you see fit.

There are even integrated AI-powered options in the lower right-hand corner of the screen that allow you to automatically adjust for things like length and tone. I chose to scroll down and review my report for my prompt. It gave me a document titled Best Practices in Generative AI for US K-12 Schools, 2023 through 2025: A Synthesis of Adoption and Efficacy.

It included multiple sections, including an overview of the rapid adoption of AI in the US educational system. It followed that up with core strategies for teacher augmentation and operational efficiency. That was followed by advanced pedagogical strategies for deep learning. Then it presented a section on foundational and supporting strategies for systemic success. Next was a synthesis of the most popular and effective strategies presented in table format, listing the top 10 strategies.

I should note that throughout the report there were clickable citations embedded so I could go to the source document and review the original materials more carefully. This provided me with even more confidence in the quality and credibility of the report. The report ended with a conclusion and recommendations, followed by a list of citations or sources used, and even the sources reviewed and not used.

At the very bottom of each report is one of my favorite features of all: a collapsible section titled "Thoughts." By clicking this, I can look at a play-by-play review of deep research's process and decision-making as it conducted the research and generated my report. It's really like looking through a transcript of an expert researcher in action. Not only did it give me additional confidence in the output content, but it also gave me ideas of how I could improve my own research process.

Step seven: sharing. Once the report was generated, I was given a few additional options. In the top right corner of the report is a share and export button. Clicking that gave me options for sharing with a link, exporting to Google Docs, or copying the contents to be pasted elsewhere. Next to the share and export button is an option titled "Create." Clicking that reveals options for generating a webpage, infographic, quiz, flashcards, and even an audio overview.

These are similar to the options available in Google's NotebookLM.

So that's how it works and what the experience is like. I have to admit, Gemini deep research impressed me, but how would I use this as a teacher?

Transition Music with Rena's Children 10:20 Integration, inspiration, integration, ideas.

Paul Beckermann 10:24 This might sound too obvious, but really, it can be useful for any situation where you want deep research. As college students, we had dedicated time to research and think academically. As educators, that time is scarce, but we still want our work to be research-based.

With deep research, you can enter a research task, let the AI do the work, and then quickly review the findings. This could be a great PLC tool for grounding pedagogical conversations in current research. It can also be great for finding research-backed instructional strategies for your classroom, or perhaps for building a research-based foundation for curricular content.

You could look for primary sources to support that content, conduct a literature analysis, find real-world applications, or further your own professional development. You know your situation best. Think about questions or wonderings that you have that require more research time than is available. Think about the times where you ask yourself, "What does the research say?" Start there.

As for students, this is clearly a tool for more advanced or older students. While you can simplify the final report with built-in filters, the ideas and connections are put together at a sophisticated level. While you would not want this to do all the hard research work for your students, it can be a nice support tool for activities where getting the research is important, but the research process itself is not the goal.

For instance, if students were planning for debates, they could use this tool to help them build their cases and anticipate counterarguments for which they need to prepare. Or maybe they're exploring scientific hypotheses and you want them to think deeply about that context. Ideas gathered through deep research might push student thinking into areas they hadn't considered or weren't aware of.

Other potential applications for students might include gathering multiple historical perspectives about a topic or time period, preparing for an argumentative essay, fact-checking current events or posts in the media, exploring personal interests with trusted sources, and even seeing examples of how topics can be researched and cited.

Again, you know your students and situation best. To get started, I'd suggest trying it out yourself. Think about something that you'd really like to research but just don't have the time to do it. Craft a good prompt, and put Gemini deep research to work.

Paul Beckermann 12:44 To learn more about today's topic and explore other free resources, visit avidopenaccess.org. Specifically, I encourage you to check out the article collection, "AI in the K-12 classroom," and, of course, be sure to join Rena Winston and me every Wednesday for our full-length podcast Unpacking Education, where we're joined by exceptional guests and explore education topics that are important to you. Thanks for listening. Take care and thanks for all you do. You make a difference.