It’s time to adopt ‘nutrition labels’ in the field of artificial intelligence

Today’s software does not come with information labels that clearly state what you are about to use in standard, simple language. What if that were the case? Regulators use nutrition labels to prevent false advertising and promote food safety. We have an opportunity to rethink health technology labeling to restore trust and help people make better choices. Imagine if the new wave of generative artificial intelligence (AI) products had clear labels to educate the public about the risks and benefits. Perhaps some of the issues surrounding the recent OpenAI board coup and governance issues could have been resolved with adequate education and warnings.

For more than 80 years, the United States has required some form of labeling on food and drugs to ensure the safety of consumables. This year marks the 33rd anniversary of the law making national nutritional labeling of foods mandatory.

Recently, tech companies have begun to realize the importance of having similar labels for fitness trackers and other products that straddle the increasingly thin line between consumer gadgets and medical devices.

Fitness trackers and their associated algorithms share many similarities with the highly regulated world of medicine and prescription drugs, including the possibility of “side effects” that produce inaccurate data in underrepresented populations, experts say . When it comes to complex products where users must make difficult decisions about benefits and risks, labels are simply essential.

What the labels might look like

Flip a box of cereal down to the nutrition labels and notice what catches your eye. Maybe it’s the amount of sodium. Or perhaps the daily percentage of protein, iron and other key nutrients.

HumanFirst CEO and co-founder Andy Coravos teamed up with researchers from Duke, Sage Bionetworks and Mount Sinai to publish an article in Nature npj Digital Medicine that examines what would be relevant to assessing risks and benefits of the use of a mobile, connected, sensor-derived digital health technology (DHT).

A mobile device can detect certain physiological measurements such as heart rate, body temperature and movement and upload data to a personal account. Connected medical devices such as mobile electrocardiograms (ECGs) intended to detect cardiac arrhythmias are becoming consumer products for home use. A consumer may value the company’s cloud health and security data encryption practices over battery life. Today, there is no good way to compare these different parameters in the same way that a nutrition label shows dietary trade-offs such as calories versus protein.

Studies suggest that simple, readable labels give consumers more confidence in the products they use, alongside a greater willingness to pay for products with trust-based labels. A UK government report reveals that three quarters of those surveyed believe it is important to introduce privacy and security labels on smart devices.

It has been proposed that companies put labels on devices to give users a clearer idea of ​​their privacy practices. This includes demanding a trusted technology brand that is only granted to products that meet a certain threshold for data rights, security and transparency. Apple has announced that apps in its store will have to include nutritional labels for privacy reasons.

In a technological culture characterized by rapid progress and disruption, there is clearly a desire to develop these types of labeling systems. We can act quickly and improve our mindset with better tools and transparent systems. In spring 2021, Duke researchers wrote Model Facts labels based on concepts developed by researchers at Google, Dartmouth and the FDA – to help doctors determine when and how to use machine learning models to clinical decisions.

Some efforts go even further down the pipeline, from algorithms to source data. Researchers from Harvard and MIT collaborated on the Dataset Nutrition Label project to measure the completeness and inclusiveness of data sets.

This year, Twillio unveiled an AI Nutrition Facts Label Generator to give consumers and businesses a more transparent and clearer view of “what’s in the box” and how their data is used.

How consumers could use health technology labels

Scenario 1: You are a consumer who may suffer from sleep apnea. While looking at a continuous positive airway pressure (CPAP) machine that fits your budget, you also want to find a lightweight device to allow travel and share data logging on your phone. You examine the “nutrition” labels on CPAP machines and choose the one that best meets your criteria.

Scenario 2: You are a principal investigator establishing a new decentralized clinical trial protocol for a psychedelic drug to treat severe PTSD, following clinical data recently published by MAPS in Nature. You are considering your REMS program and looking to evaluate the risks and benefits of various remote monitoring products. For some patients, a wearable device that constantly collects their data can be frightening and they fear being constantly monitored and tracked. On the other hand, a well-designed wearable device can provide a sense of safety and security to these trial participants.

Scenario 3: You are an executive in a biopharmaceutical company. You want to collect metrics that are important to patients, such as I could climb the stairs Or I slept better. Collecting quality of life and performance outcomes using sensors (and using labels to assess quality ones) could change the way clinical trials are conducted. More than 130 pharmaceutical and biotechnology sponsors have collected digital endpoint data to submit to regulators. The use of labels could help move towards a more practical approach to drug discovery and development.

The HumanFirst Atlas is primarily designed for pharmaceutical and biotechnology teams to design and deploy more effective precision measurements in clinical trials. The company is interested in user-friendliness, privacy and data security. This information could begin to be used to develop health technology labels that would be more widely leveraged by consumers and doctors for personal care. Although there remains a need for education to empower consumers in their decisions, the first step is to centralize information in an objective and impartial manner to give them more choices.

By evaluating new digital tools, we can learn lessons from the pharmaceutical and food industries to increase trust in health technologies. We can learn to document and share adverse events, such as detecting and reducing algorithm bias, and better information can help consumers make decisions about how best to use products. Well-designed labels can contribute to all of these aspects to build a more human-centered healthcare infrastructure.