Artificial intelligence for diagrams and graphs

I think that AI can help boost creativity and design diagrams faster. In addition, if you’re not skilled in drawing, instead of drawing a diagram or graph by yourself, you can use speech to draw and create one.  

AI can also generate and interpret diagrams or graphs, read them, and understand what they represent, including relationships between nodes. It can help users determine what to modify on the diagram and what nodes or objects to add or remove.

For example, you can select a part of a graph and ask an AI assistant if it’s the best way to represent an idea. Or maybe there is a better option for that?

So, AI can help you generate more ideas, accelerate your work with creating diagrams or graphs, and eventually help you interpret them or make them more accurate.

Frank Joublin,

Principal Scientist at Honda Research Institute

One of the biggest benefits of using AI for end users is boosting productivity. Everything is related to productivity.

The complexity of diagrams increases, so you can’t build them in the traditional way by using drag and drop and designing everything.

Looking at the complexity of my projects that involve building graphs that handle conversations with hundreds of steps, it’s not realistic anymore to develop them using old methods.

Another thing is that when I design a diagram, I would like to get help from an AI model to tell me if I should design it this way or that way or if I have the following sinks in my diagram to take care of.

I think it’s beneficial to have an AI model that understands my data, whether they are conversation transcripts, sales orders, HR documents, etc.

Daniel Balaceanu,

Chief Product Officer at DRUID AI

For sure, AI in the future will generate graphs more intensively than now. This is a trend that will stay with us.

Let’s take an example. For every domain where you need a kind of map visualization, it’s impossible to describe it using words. By seeing a map, you can show something on it, point to it, interact with it, and highlight some aspects. This is something you can’t do using only words. You can’t describe complex structures only with words.
 
So, I think we’ll have more and more AI solutions that generate graphs because they are a great means to represent complex data.

Frank Joublin,

Principal Scientist at Honda Research Institute

I think the most significant obstacles are security, data privacy, people's acceptance, and the ethical dimension of AI.  
 
The security aspect is nothing new—there are still companies that would use on-premise solutions rather than leverage cloud services offered by established companies such as AWS, Google or Microsoft. AI is no different, early adopters will move in, others will join later. Some won't show up at all.  
 
Adding to that AI, some people might have ethical doubts because they're unsure where the data comes from and if it includes people's personal details or intellectual property.
 
So, as a consequence, they still determine if they should use AI solutions trained on such data of unknown origin. In addition, they are afraid of what will happen to their data.  
 
For me, it's a significant challenge because I build my product as a social platform and potentially offer it to 200 million creators or writers. It means I must address these ethical concerns.  
 
Users should have the right to choose whether to opt in or opt out of AI features in products they use. If they don't want it, AI models should not consider their data.

Anubhav Srivastava,

CEO and Co-founder at VocaTales

Data security concerns all my projects, and I must take care of it, especially in all new technology I use. Each technology should be embeddable in my product. It means that I must host it and manage the entire data flow.

Most of our customers leverage technology from the cloud. We also manage cloud services. However, then we audit everything and guarantee that we are the only ones that hold our customers' data.

And if we use third-party services like OpenAI for Microsoft subscriptions, we must sign data privacy contracts. And we work only with already accepted providers in the enterprise sector like Microsoft or Google.

As for other challenges, the performance of AI models is not a big obstacle. From my experience they usually respond in 1-2 seconds.

In complex implementations, you can onboard the respective hardware or ask Microsoft or any other provider for special subscriptions.
 
The market is moving in that direction to have faster response time and I see it especially in the US market. I have talked with several banks, and they're starting to apply hardware just for multiple experiments with generative AI.  
 
So large businesses are getting prepared for broader use of AI solutions.

Daniel Balaceanu,

Chief Product Officer at DRUID AI

If a company wants to have AI on-premises, it's the point where things become complicated.

My observation is that if you want to train a model, you need huge amounts of data, infrastructure, and experts like data engineers and devops specialists. But that's a bit of a problem. For me, it's a job of its own.

Francois Vanderseypen,

CEO at Orbifold Consulting

The first constraint that comes to mind is data protection. One day, we might have to develop LLM to protect our sensitive data.
 
The second is the large language models' response time. If it’s too long, you will get bored and stop using them.
 
And, of course, users must also cope with AI's well-known problem - the hallucinations that cause you can't be sure that the AI outcome reflects the truth.

Another problem is that when LLMs are updated, they may behave differently. As a result, you might have to redefine your prompts because the models don't understand them the same way as their previous version.

Frank Joublin,

Principal Scientist at Honda Research Institute

When it comes to “AI-powered solutions” there are some challenges. The typical ones are resources – not only the budget but also the lack of experienced specialists. More specifically, finding those suppliers that are trustworthy and substantially better than in-house developers. The problem is that it’s tough to separate trustworthy vendors' because wide-scale AI usage is relatively new, experts are still figuring it out and meanwhile there’s a lot of boasting about skills.   
 
Another interesting observation is the push for integrating AI into software solutions.  And whilst AI is the future, at its current stage, it shouldn’t be the most pressing development on every product roadmap. The most pressing items should be related to solving customer problems and with that comes getting your basic right. So, a potential bigger obstacle I’m noticing in the market is companies being forced to integrate AI into their tools, and by doing so, they might not be focusing on what they should do.  
  
As an example, having a Chat GPT prompt builder is very cool to have on your platform, we all agree with that. But it’s extra functionality, once you get the essence of your solution right.

Peter Theunis,

Co-founder of Erisna

Surely, the obstacle is a lack of data to train models. Large language models need extensive data to be helpful. It's a fundamental limitation. I think the big problem with our model is that we don't have enough training data and good, simple examples of beautiful diagrams to train it and make something worthwhile.  

So that's why we feel that we should wait and see. Maybe in a few years, there will be models that can do a lot more with a lot less. But right now, they need a lot of data to be useful.

There's also the privacy concern, at least for some companies. If you offer a tool, you can scare some people who are worried that if they start using anything from your company, you'll start using their data to train your models.

If you offer a tool that does anything automatic, you must be sensitive to the customer and make sure they know you're not trying to take any of their data at the same time.

Simon Sarris,

Software Developer at Northwoods Software Corporation

I don’t think that graphical interfaces will disappear like that because you can’t explain the same thing using plain text or speech instead of a visual interface – it would take too much time.  

A lot of people think visually and represent concepts using boxes and links. Written and spoken language is suitable for many aspects, but not for all of them. There are domains where graphical representation is needed because language cannot describe thoughts and concepts. I think the graphical representation will stay, and the audio interfaces will coexist.

Frank Joublin,

Principal Scientist at Honda Research Institute

It's a tricky question. It sounds like we no longer need graphical interfaces, such as traditional tables, charts, and graphs. And that all of us will stop using any other kind of interaction except text and audio.
 
It doesn't work like that.  
 
The models themselves evolve. At first, Chat GPT provided just text input. Now, I can upload images and get output based on them.

So, in my opinion, the models will evolve to offer a multimodal experience, engaging users not only through text or audio but also with other forms, including diagrams.

I believe I could soon form a prompt like that: I want to create a conversation graph that allows customers to be onboard for a credit application. My credit application should capture the following parameters. It's integrated with SAP with three back-end APIs. Can you draft the diagram for me, please?

I think general domains will be involved in developing AI solutions in this direction, specialized models will appear, and existing business applications will start having such functionalities.

Daniel Balaceanu,

Chief Product Officer at DRUID AI

I think that graphical interfaces make so much sense and will develop in the future.

Take a look at the younger generations, such as Generation Z and Alpha. Those kids love video games that are based on graphical interfaces. There’s no kid that doesn’t spend time watching Minecraft videos.

Looking at how they think, they process visual elements much faster than written text.
It doesn’t mean that text will disappear because you need text to create all the visuals. You must write all the details that you need first.

But graphical interfaces will continue being important because they help people think and act faster.

Anubhav Srivastava,

CEO and Co-founder at VocaTales

In my world, despite many advancements, people lean towards visuals and graphical representations to form or showcase conclusions. Those visuals then get supported with a narrative. The narrative is very important, however at this point, it seems that visualizing helps people understand and form conclusions. Typically, for any graphical representation you’d have numerous data points, which could easily be shown in tabular, text or audio format, yet as a society we seem to be sticking to graphical interfaces for important items
 
Let's take data lineage or data mapping. The idea behind it is how can I showcase a data flow in an appealing way so that the person looking at it can understand it and make quicker decisions. There’s no reason we couldn’t explain this through text or audio yet as end-consumers we prefer a visual interface.
 
I don't think it will change anytime soon.

Peter Theunis,

Co-founder of Erisna

Graphical interfaces, like BI dashboards, will remain. In any case, people will always like bar charts. Text will not replace charts, and a paragraph can't replace mind maps. So, text and graph interfaces are complementary. However, I suspect that Chat GPT will be able to create a dashboard at some point.

Francois Vanderseypen,

CEO at Orbifold Consulting

Even visual data can lie. It still can hallucinate. But because people's attention span is shorter and shorter, we must see things faster.   
 
How can you spot errors in a document that is pure text? We must visualize it first to make it easier to consume and analyze. That's exactly what we are doing on VocaTales through storyboarding.

Anubhav Srivastava,

CEO and Co-founder at VocaTales

I always have this example in the back of my mind with machine learning. One of the first very successful machine learning applications ever was PayPal in the early 2000s.
They built this very early machine learning system to detect fraud. But one characteristic of payments is that you're very fuzzy about what is fraud and what isn't.

So, a human must review the results. If you're crunching hundreds or thousands of transactions and can get a view at the end, you should have a chance to order it or make it more readable. So that person can look at the results and say: those seven transactions look bad.

So, visualization is essential for a person who has to review all of those outcomes.

Simon Sarris,

Software Developer at Northwoods Software Corporation

Having good LLMs and knowledge on how to use them effectively for diagram creation is, in my opinion, the winning part.  

Prompt builders can theoretically accelerate building diagrams. However, in practice, when I apply them in production, I sometimes realize I don’t have much control. I write a text describing how the model should behave, but then it doesn’t behave how I expected. It’s difficult to troubleshoot and understand what happened. For instance, why did the conversation move on that path, or what are the paths?  

So, we need tools to handle that.
 
Prompt builders can also create complexity that is difficult for a user to handle. Suppose a prompt builder generates a diagram with 80 steps. In that case, you must take over this outcome and adjust it after it finishes. So, instead of simplifying, the model creates complexity that the user must handle now.

Daniel Balaceanu,

Chief Product Officer at DRUID AI

There's a huge interest in AI, but businesses need guidance in leveraging it.  

You can't keep up with everything that is happening—the framework, AI, language models, the difference between some types of diagrams, whether you choose Meta or Microsoft.

So business owners need guidance in the first place to make sense of all of the things that are going on. Requesting augmented generation is usually the second thing.  

For instance, people have PDFs, knowledge graphs, or PostgreSQL databases and want to know what it takes to use natural language to talk to your database. That's the typical case.

People or companies do that partly to attract new employees and show that their firms are up to date with tech news.

So, my role is often to offer something like a guide in the first place because I know how you can go somewhere and what it takes. Where you can search for this type of model or tools that you can utilize with projects for your customers.

Francois Vanderseypen,

CEO at Orbifold Consulting

There are some areas that I would happily hand over to AI. However, helping clients create custom applications and understanding their business processes is a task for people.  

What I do is more than coding. I can't imagine an AI that sits in a meeting for long hours and listens to political clashes between different departments within a big organization. And that's also what I do. Let's remember that AI is a technology and not a human being. It can't replace humans in dealing with all the complexity of corporate life.  

So, I'm not afraid that AI will replace my business consultancy role soon. And I have a similar opinion as for custom development services.

Another thing is that I don't want AI to handle me with the creative part of my work because I enjoy it.  

Business is about dealing with people, talking to them, and understanding their ideas or visions. There are rational and irrational aspects, social dynamics, communication, trust, and other vital elements.

Business owners talk to each other because they believe it results in some wisdom and is more meaningful than asking Chat GPT.  

Sitting at a table and having dinner sometimes is just as important for business as developing JavaScript.

So, business owners should invest in good relationships and kindness.

Francois Vanderseypen,

CEO at Orbifold Consulting

Currently, companies are using SAP and mobile applications. But should I say those companies are run by SAP or Power BI reporting tools only because they use them daily for making decisions?

I would say no.  

Those tools are used by humans and humans drive the business.

Artificial intelligence is a catchy and provoking name, but that’s it. It’s not that robots and AI will replace humans. The humans who won’t use AI will lose their jobs.

Daniel Balaceanu,

Chief Product Officer at DRUID AI

It has already happened in some areas and is a risk no one should underestimate. Nevertheless, it is the responsibility of all of us to try to develop solutions that have a positive impact on society.  
 
For instance, it can be used in domains without humans involved because of unrealistic costs. I think of personal teachers for every child, personal aids for elderly people, or personal moderators of your social media interactions. The idea is to use AI to complement and collaborate instead of replacing.
 
So, it can replace some jobs. However, AI will initially help many specialists do their work more efficiently.

Frank Joublin,

Principal Scientist at Honda Research Institute

One thing I remember from one of the conferences I took part in was a speaker from Microsoft who said that it will take time for AI to reach performance of human workers in many domains, however jobs like data scientists are among the few jobs that could likely be performed by AI at a great extent.

Of course, we must consider that it has been said by a Microsoft employee who works closely with OpenAI. So, he might have wanted to convince the audience that this is the future.

So, now AI helps people rather than replaces them, and we can hope it will stay like this for a long time.

Antonello Ceravola,

Principal Scientist at Honda Research Institute

The dust has to settle. I think there is a crazy amount of uproar about AI everywhere.  
 
Just this week, I interviewed around 100 sixteen- to twenty-two-year-olds, asking them about their point of view on AI and where they think it stands. Almost everyone said that AI is a great addition to enhancing human creativity. But, in the end, the real revolution and changes must come from the human brain.  
 
As the dust settles slowly and steadily, AI will become an integral part of everything that we do.  
 
It already has become. However, when it comes to the kind of work involving human creativity, imagination, thought process, experimentation, and experiences, its results will still trump what some AI-generated work produces. All the more when humans collaborate to create and transform the future for the better.

Anubhav Srivastava,

CEO and Co-founder at VocaTales