“Code Interpreter is a good thing and it’s helpful in an educational setting as it makes coding in the STEM fields more accessible to students,” said Gangqing “Michael” Hu, assistant professor in the Department of Microbiology, Immunology and Cell Biology at the WVU School of Medicine and director of the Bioinformatics Core. “However, it doesn’t have the features you need for bioinformatics. These are technical issues that can be overcome. Future developments of Code Interpreter are likely to extend its use to many fields such as bioinformatics, finance and economics.”

Since its release in December 2022, the popular artificial intelligence chatbot ChatGPT has gained the attention of businesses, educators and the general public. However, it didn’t quite live up to the needs of people working in biomedical research including bioinformatics — the field where computer science meets biology — who eagerly awaited OpenAI’s Code Interpreter plugin hoping it would fill the gaps.

Hu and his team put Code Interpreter to the test on a variety of tasks to evaluate its features. Their findings, published in Annals of Biomedical Engineering, show the plugin breaks down some of the barriers, but not all of them.

For example, people without a science background will have an ease of access to coding, or computer programming, with Code Interpreter. Hu said it’s also cost-effective and sparks a curiosity for students to explore data analysis and boosts their interest in learning. He points out, though, users will need to understand how to interpret data and recognize whether the results are accurate and know how to interact with the chatbot.

Bioinformaticians rely on precise coding, computer software programs and internet access to store, analyze and interpret biological data such as DNA and human genome used for advancements in modern medicine.

Despite the need for improvements specific to bioinformatics, Hu said, Code Interpreter helps users determine whether a response is accurate or if it is a fictitious answer presented with confidence, known as a hallucination.

“People know that ChatGPT can do many impressive things, but it is not good at providing a citation or reference to support its answer. If it is asked about the source to support the claim of a response, it may start to make up references,” Hu explained. “Code Interpreter provides a solution to minimize hallucinations. For questions that can be addressed through coding, the code itself serves as the source or citation. That is a significant step forward.”

Working with Hu were Lei Wang, a postdoctoral fellow in the WVU Department of Microbiology, Immunology and Cell Biology; Xijin Ge, of South Dakota State University; and Li Liu, of Arizona State University.

The team found positive results in Code Interpreter’s ability to convert data to charts and graphs.

Suggestions for upgrades to Code Interpreter include internet access for downloading genome data, installation of software specific to bioinformatics, expansion of storage capacity and support for additional programming languages. In addition, researchers found a need for privacy and security applications to comply with regulations such as HIPAA.

In testing data analysis, they discovered several limitations. The plugin supports only one computer program, Python, and few of its software packages are dedicated to bioinformatics. In addition, it doesn’t allow access to internet data and lacks the capacity to work with large files.

“It allows for 100 megabytes or so, but the files we’re handling are at a gigabyte level,” Hu said. “Also, it doesn’t support parallel processing needed for large datasets which results in slow performance.”

Hu said that while he anticipates more upgrades for Code Interpreter, he plans to help students learn more about the advantages of the current plugin.

“In my class next spring, I plan to introduce this plugin to help students learn about data visualization,” Hu said. “AI is a fast-moving field. I hope by that time OpenAI may overcome some of the limitations so it can be used for a broad range of bioinformatics coding.”

Earlier this year, Hu led another study to prepare high school and college students to harness the power of ChatGPT by learning more about coding. The process employed OPTIMAL — Optimization of Prompts Through Iterative Mentoring and Assessment — to improve communication with a chatbot.

In the long run, Hu said he will continue to monitor and test new AI programming and features.

“As new products develop, I’ll just keep going,” Hu said. “There are certainly many other innovative uses awaiting to be discovered.”

Humans are generally good at whole-body manipulation, but robots struggle with such tasks. To the robot, each spot where the box could touch any point on the carrier’s fingers, arms, and torso represents a contact event that it must reason about. With billions of potential contact events, planning for this task quickly becomes intractable.

Now MIT researchers found a way to simplify this process, known as contact-rich manipulation planning. They use an AI technique called smoothing, which summarizes many contact events into a smaller number of decisions, to enable even a simple algorithm to quickly identify an effective manipulation plan for the robot.

While still in its early days, this method could potentially enable factories to use smaller, mobile robots that can manipulate objects with their entire arms or bodies, rather than large robotic arms that can only grasp using fingertips. This may help reduce energy consumption and drive down costs. In addition, this technique could be useful in robots sent on exploration missions to Mars or other solar system bodies, since they could adapt to the environment quickly using only an onboard computer.

“Rather than thinking about this as a black-box system, if we can leverage the structure of these kinds of robotic systems using models, there is an opportunity to accelerate the whole procedure of trying to make these decisions and come up with contact-rich plans,” says H.J. Terry Suh, an electrical engineering and computer science (EECS) graduate student and co-lead author of a paper on this technique.

Joining Suh on the paper are co-lead author Tao Pang PhD ’23, a roboticist at Boston Dynamics AI Institute; Lujie Yang, an EECS graduate student; and senior author Russ Tedrake, the Toyota Professor of EECS, Aeronautics and Astronautics, and Mechanical Engineering, and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL). The research appears this week in IEEE Transactions on Robotics.

Learning about learning

Reinforcement learning is a machine-learning technique where an agent, like a robot, learns to complete a task through trial and error with a reward for getting closer to a goal. Researchers say this type of learning takes a black-box approach because the system must learn everything about the world through trial and error.

It has been used effectively for contact-rich manipulation planning, where the robot seeks to learn the best way to move an object in a specified manner.

But because there may be billions of potential contact points that a robot must reason about when determining how to use its fingers, hands, arms, and body to interact with an object, this trial-and-error approach requires a great deal of computation.

“Reinforcement learning may need to go through millions of years in simulation time to actually be able to learn a policy,” Suh adds.

On the other hand, if researchers specifically design a physics-based model using their knowledge of the system and the task they want the robot to accomplish, that model incorporates structure about this world that makes it more efficient.

Yet physics-based approaches aren’t as effective as reinforcement learning when it comes to contact-rich manipulation planning — Suh and Pang wondered why.

They conducted a detailed analysis and found that a technique known as smoothing enables reinforcement learning to perform so well.

Many of the decisions a robot could make when determining how to manipulate an object aren’t important in the grand scheme of things. For instance, each infinitesimal adjustment of one finger, whether or not it results in contact with the object, doesn’t matter very much. Smoothing averages away many of those unimportant, intermediate decisions, leaving a few important ones.

Reinforcement learning performs smoothing implicitly by trying many contact points and then computing a weighted average of the results. Drawing on this insight, the MIT researchers designed a simple model that performs a similar type of smoothing, enabling it to focus on core robot-object interactions and predict long-term behavior. They showed that this approach could be just as effective as reinforcement learning at generating complex plans.

“If you know a bit more about your problem, you can design more efficient algorithms,” Pang says.

A winning combination

Even though smoothing greatly simplifies the decisions, searching through the remaining decisions can still be a difficult problem. So, the researchers combined their model with an algorithm that can rapidly and efficiently search through all possible decisions the robot could make.

With this combination, the computation time was cut down to about a minute on a standard laptop.

They first tested their approach in simulations where robotic hands were given tasks like moving a pen to a desired configuration, opening a door, or picking up a plate. In each instance, their model-based approach achieved the same performance as reinforcement learning, but in a fraction of the time. They saw similar results when they tested their model in hardware on real robotic arms.

“The same ideas that enable whole-body manipulation also work for planning with dexterous, human-like hands. Previously, most researchers said that reinforcement learning was the only approach that scaled to dexterous hands, but Terry and Tao showed that by taking this key idea of (randomized) smoothing from reinforcement learning, they can make more traditional planning methods work extremely well, too,” Tedrake says.

However, the model they developed relies on a simpler approximation of the real world, so it cannot handle very dynamic motions, such as objects falling. While effective for slower manipulation tasks, their approach cannot create a plan that would enable a robot to toss a can into a trash bin, for instance. In the future, the researchers plan to enhance their technique so it could tackle these highly dynamic motions.

“If you study your models carefully and really understand the problem you are trying to solve, there are definitely some gains you can achieve. There are benefits to doing things that are beyond the black box,” Suh says.

This work is funded, in part, by Amazon, MIT Lincoln Laboratory, the National Science Foundation, and the Ocado Group.