Ansys Human Explains Why AI Is Handling Some of the Firm’s Tech Support

VP of Customer Excellence: AnsysGPT will provide engineers high-quality tech and simulation support 24/7.

In late July, Ansys announced its new generative AI tool, AnsysGPT. The AI-powered virtual assistant is multilingual, conversational and relies on natural language processing (NLP) to interact with engineers. Like other human technical agents, the goal of AnsysGPT is to provide 24/7 technical support for engineers on any issue related to Ansys simulation or other solutions. Vice president of Customer Excellence, Anthony Dawson, sat down with engineering.com to discuss how this change could impact Ansys users.

An example of the AnsysGPT interface and interactions with its AI-powered virtual assistant. (Image: Ansys.)

An example of the AnsysGPT interface and interactions with its AI-powered virtual assistant. (Image: Ansys.)

He explained that previously, individuals had to wait for a limited number of human agents to assist with technical inquiries during the appropriate business hours. Though the tool is only available as a beta launch for select customers, the hope is that engineers will get the help they need, when they need it, after a full release to the public in early 2024.

Training an Engineering AI Assistant for Simulation

AnsysGPT was developed using ChatGPT through the Microsoft Azure OpenAI Service. The tool was trained using exclusively Ansys data, ensuring all answers are customized to Ansys services, solutions and customers. The training dataset included Ansys courses, blogs and technical documentation. It has also learned from how-to videos to ensure it is an effective first point of contact for Ansys-specific technical issues.

“AnsysGPT is fundamentally different from other LLMs [large language models] in a few ways—one being how we approached training data,” explained Dawson. “We fed AnsysGPT a strict diet of verified Ansys resources. The result is a knowledge engine that combines subject matter expertise usually spread across multiple engineers or engineering teams into one platform. Our customers aren’t looking for general information; they need Ansys-specific answers, and they need to know they can trust the responses.”

This entire training dataset is already available to the public, but it can be challenging to sift through it when searching for an answer to a technical question. With AnsysGPT, engineers can efficiently find the appropriate source material to support their queries with a conversational interface.

AnsysGPT also relied on human expertise as part of the training workflow. Ansys took advantage of its internal customer support team to assist with prioritizing the different datasets during training. The existing team already knows which documents are most frequently used by engineers and which are the most useful for its clients. All this human metadata was factored into the prioritization process and altered the hierarchy of sources referenced in generated answers for the tool.

Usually, conventional tools like ChatGPT will use the inputs and outputs generated through user interactions to further train the model. According to Dawson, Ansys will not be using the prompts or responses of users to further teach the tool. The motivation behind this restriction is to protect customers’ data and ensure that users feel comfortable taking advantage of the assistant. Ansys knows that its customers work with a lot of proprietary data and it wants those users to feel comfortable using the tool, knowing their prompts, data and responses will not be used to further improve the solution or aid other users. So, instead of using inputs for training, Ansys relied on beta user feedback to continually improve the tool.

Can AI Reproduce Human-Generated Technical Solutions?

To date, Ansys has not been able to directly compare the performance of its human agents with the AI virtual assistant. For now, the company is observing the differences between the answers generated by AnsysGPT and the ideal answer generated by a customer support team member. Dawson indicated that so far, the beta testing clients have been satisfied with the tool and have provided positive feedback related to their interactions.

In describing the differences between the AI and human agents, Dawson added, “To a certain extent, it’s always going to be apples and oranges.” However, he did not offer any statistics related to how often the generated answers align with their best response. Ideally, this information would be shared when the tool becomes available to the public.

“AnsysGPT can accelerate the process of our customers asking questions and getting answers, finding helpful resources or identifying an issue and receiving a solution,” said Dawson. “And like other generative AI tools, it can serve up resources with technical information that can be a good reference point for more advanced answers. We’ve all had experience searching for a needle in a haystack. It can be frustrating and time consuming, and it doesn’t have to be anymore.”

The tool also directs users to standard Ansys support for inquiries that require complex, tailored solutions. So, it seems that instead of having AnsysGPT completely replace the customer service department, the goal will be to streamline the process of seeking assistance. This will enable human agents to focus their expertise on more complex technical issues.

What Else Can a Virtual Assistant Accomplish?

In considering Ansys expertise, it is exciting to think of a future when a virtual tool like AnsysGPT could run simulations. For example, in “Star Trek: The Next Generation,” users could simply describe a simulation and the computer automatically processed the job; no input was necessary. Such a tool would significantly accelerate the simulation process and allow nontechnical experts to run and process simulation data.

When probed about this possibility, Dawson stated that he was unsure whether this would be a future goal for Ansys. For now, the AI assistant is only one additional step in the company’s strategy to integrate more AI into its offerings and solutions.

However, Dawson did add that “there are definitely some exciting synergies between AnsysGPT and PyAnsys. Through PyAnsys, a unified API customization language for all Ansys products, it’s possible future iterations of AnsysGPT could directly interact with users as they set up their simulations. It’s too early to say what AnsysGPT’s additional capabilities will look like or when they’ll arrive, but we see a bright future for this technology.”

Although AnsysGPT is not currently capable of running simulations itself, the accessibility of its assistance can still improve simulation speed and efficiency. Depending on the inquiry, engineers can now receive technical support that is highly specific for Ansys solutions at any time, allowing questions to be answered significantly faster. By reducing the time to a solution, engineers can improve the efficiency of their simulation—but only if they are regularly derailed by technical issues. If no problems arise, the tool doesn’t assist with runtimes or simulation setup.

Ansys Adds to Its Growing AI-Enabled Portfolio

AnsysGPT adds to the long list of AI-enabled solutions offered by the company, including computational fluid dynamics (CFD) solvers that use AI to optimize turbulence, structural tools that use AI to predict computational spend, and tools that use AI for design optimization to build reduced order models.

However, one crucial question remains: beyond the beta testing, what is the preference of the average engineer? Do we want to interact with AI to solve these technical issues? Is there still a preference for interaction with a human agent? I think the future will likely continue to see a mix of both. AI assistants like AnsysGPT will be used in the way the company intends: to offload simple technical requests. On the other hand, human agents will be prioritized for more complex and sensitive issues that truly require ingenuity to solve.