As featured by Engineering.com
Engineering leadership must embrace AI in the innovation process.
Research and development (R&D) teams don’t have it easy. They are facing shrinking workforces, budgets and development times. Meanwhile, the ideas, products and solutions they provide are becoming more complex, multidisciplinary and competitive.
“The traditional workflows engineers use to innovate and solve problems include trial and error, brute force, Kaizen and more,” says Ameet Bhattacharya, CTO of IP.com. He explains that inventive problem-solving (IPS) methodologies enable systematic ways to innovate. They often include studying inventions, across multidisciplinary domains, to find solutions to similar problems.
However, Bhattacharya suggests IPS tools — by themselves — are not capable of helping R&D teams navigate the complexities of their designs in a fast manner. “How do you rapidly share ideas and thoughts as part of the team? How do you look up all the solutions out there and come up with an idea that’s novel and patentable? How do you quickly do that novelty assessment?”
James Durkin, managing director of Product Management at IP.com answers these questions with two letters: AI. “Artificial intelligence will have a dramatic effect [on R&D] from initial ideation all the way to the patenting phase. It will only accelerate and come to a point where it’s a standard operating procedure to use generative AI in every part of the engineering workflow.”
If engineering leadership and R&D teams don’t act, they risk falling behind the AI revolution. But how should they proceed?
AI will improve the productivity of R&D
AI is bound to change up every aspect of the business, and R&D is no exception. The obvious way things will change include the new products engineers and R&D teams come up with. But from a leadership perspective, it’s also important to anticipate how AI will change the way R&D engineers do their jobs in the first place.
“The performance of an engineer can be measured by the number of usable or unique ideas they come up with and the patents they produce,” says Durkin. He suggests that R&D teams can use AI to increase their productivity by speeding up the number of ideas they generate. “Software needs to be an accelerator to traditional methodologies. Not replace them, just augment them to help engineers develop faster.”
But this only raises more questions: How do R&D leadership and engineering teams overcome the challenges associated with AI such as training, hallucinations and intellectual property (IP) protection? How do they tackle the legal implications of using AI in the patent process? And how do they marry AI to traditional problem-solving methodologies?
Bhattacharya suggests they use a blend of semantic and generative AI technologies that are guard-railed by IPS methodologies and supported by a wealth of data insights contained within technical literature, research and patents. In other words, they should try IP.com’s innovation management and ideas generation solutions, but more on that later.
The types of AI that can be useful to R&D leadership and engineers
First, R&D leadership must determine the ways AI can improve the innovation process and which flavors of the technology are most suited for the task. Bhattacharya and Durkin focus on three types of AI technology in their discussion:
Large language models (LLM): deep learning algorithms that are used for natural language processing (NLP) tasks.
Semantic search: a subset of LLMs that process the text and meaning of a prompt to retrieve results from the data it was trained on.
Generative AI: another subset of LLMs that process the text and meaning of a prompt to produce an original response based on the data it was trained on.
Though these three AI technologies are similar, it is important to note their distinctions. “Generative AI produces data similar in characteristics from what it was trained on,” says Durkin. “It doesn’t have a good understanding of what it means. [Its response] is based on probability and that’s how hallucinations happen.”
Semantic search, on the other hand, doesn’t create its own data, and thus cannot hallucinate. It instead finds and serves the data (from the data pool it was trained on) it thinks best matches the prompt.
Durkin best summarizes the potential risks of AI in an R&D setting. “The biggest risk of semantic AI is that you read extra data you didn’t need. The biggest risk of generative AI is that you make up things, so the risks are infinite but manageable with the right technology.”
Either way, the output is as true and good as the data used to train the model. Thus, to reduce these risks, it is key to get the best training data available. “Data is important because if it isn’t trained properly that’s where you get biases,” confirms Bhattacharya. “But there is also a chance that the data the algorithm serves has no relation to the prompt it was fed.”
Bias is another concern and limitation of AI technology. “Quality directly affects the performance of the model,” says Bhattacharya. “Biases and inaccuracies in data are critical.” He adds that the quantity of data is also important. With too little data, the algorithm can become overfit or suffer from gaps in knowledge. “Poor quality [training data] gets erroneous outputs and generalizes data. AI can learn incorrect patterns.”
How to incorporate AI into the innovation process
The better R&D leadership and engineers understand AI, the better they can optimize how they use, train and prompt it. For instance, humans can tweak and modify their prompts based on their knowledge of an AI’s workings and its training data. These tweaks aim to optimize for the outputs. This process is called meta prompting.
Durkin suggests that by training semantic search and generative AI algorithms on proven IPS methodologies, technological literature and patents, they can be used in conjunction, in a process called IPS meta prompting.
“Users might not know what kind of inputs to put into semantic AI to get the best outputs,” he says. “Generative AI can help … It lowers the entry barrier. Young engineers don’t have to be trained in these IPS methodologies. It’s easier to perform these methodologies and democratizes them.”
Consider an engineer who needs to search the patent literature to ensure their idea is novel. They could go straight to the semantic algorithm, describe their idea and hope the tool finds every patent that is relevant. But what if they instead described it to a generative AI algorithm trained on similar data? That could then produce an IPS meta prompt able to get the semantic algorithm to retrieve the best list of patents possible.
“Generative AI helps to describe the problem, then plugs that into the semantic AI to find the most relevant results to what you’re looking for,” says Durkin. “The generative AI hands it’s output to semantic AI that then gives a list of what has been done before to help guide the problem solving.”
“Understanding the prior art helps guide the solution development process to not infringe on existing IP while also not reinventing the wheel,” explains Bhattacharya. “You learn from past mistakes and successes and legal risks with the solutions you pursue. You find emerging trends in the industry. Getting that list of patents is important.”
The risks of using AI in R&D
Some R&D specialists new to AI might be asking, ‘why not have the generative AI play a more active role in the idea, or patent, creation?’ Besides the risks of hallucinations creeping in this way, there are legal quandaries.
In late 2023, Britain’s Supreme Court ruled that machines cannot be considered an inventor under current law. Thus, in Britain (and likely other jurisdictions) to achieve a patent, organizations must ensure all those that came up with the idea are human.
Secondly, as the generative AI is trained on current art, there is a chance its results infringe, or even plagiarize, previous work.
Durkin says AI best helps R&D initiatives with structured ideation. He adds, “generative AI helps formulate the ideas. It won’t fill all the gaps. You still need humans to help there.”
AI trends and opportunities for R&D and engineering
IP.com offers a software as a service (SaaS) model to access its IQ Ideas+ and InnovationQ+ technologies. The former uses AI to help R&D teams generate and refine ideas to maximize their potential early in development. The latter uses AI to perform patent searches and analytics to improve the ROI of the innovation lifecycle.
Bhattacharya explains that with these tools, engineers can “go through the ideation tools, put in prompts, upload documents about the prompt, and the systems being improved on. Then after that they go through the interactive data.”
He adds that “collaboration is another big area. Seamlessly work as a team and share ideas and solutions and validate that.” In other words, these tools help to bring multidisciplinary knowledge into the idea process to help propose holistic solutions. Bhattacharya says, “this is very challenging without AI.”
He adds that the future of IP.com’s technology is very interesting. Soon organizations will be able to train AI on proprietary information, data and designs. “You create your own AI that is separate and distinct from what’s available out there,” says Bhattacharya. “But it is trained on your own data. It’s a custom implementation of AI that won’t talk to other implementations of AI.”
It goes even further. Bhattacharya envisions an AI future where R&D teams can make digital twins of an object just by prompts. They could then test and iterate on their designs using augmented and virtual realities (AR/VR). They could even automate tasks like making CAD or the simulation process. This way R&D teams can focus on the more interesting parts of the design process.
Clearly, AI will make the future of R&D bright, but it will also require leadership to get on board quickly. Testing a proof of concept via an IP.com free trial might be a good first step.
