fbpx
Connect with us

The Conversation

Human differences in judgment lead to problems for AI

Published

on

theconversation.com – Mayank Kejriwal, Research Assistant Professor of Industrial & Engineering, of Southern California – 2024-05-14 07:14:06

Bias isn't the only human imperfection turning up in AI.

Emrah Turudu/Photodisc via Getty Images

Mayank Kejriwal, University of Southern California

Many people understand the concept of bias at some intuitive level. In society, and in artificial intelligence systems, racial and gender biases are well documented.

Advertisement

If society could somehow bias, would all problems go away? The late Nobel laureate Daniel Kahneman, who was a key figure in the field of behavioral economics, argued in his last book that bias is just one side of the coin. Errors in judgments can be attributed to two sources: bias and noise.

Bias and noise both play important roles in fields such as law, medicine and financial forecasting, where human judgments are central. In our work as computer and information scientists, my colleagues and I have found that noise also plays a role in AI.

Statistical noise

Noise in this context means variation in how people make judgments of the same problem or situation. The problem of noise is more pervasive than initially meets the eye. A seminal work, dating back all the way to the Great Depression, has found that different judges gave different sentences for similar cases.

Worryingly, sentencing in court cases can depend on things such as the temperature and whether the local football team won. Such factors, at least in part, contribute to the perception that the justice system is not just biased but also arbitrary at times.

Advertisement

Other examples: Insurance adjusters might give different estimates for similar claims, reflecting noise in their judgments. Noise is likely present in all manner of contests, ranging from wine tastings to local beauty pageants to college admissions.

Behavioral economist Daniel Kahneman explains the concept of noise in human judgment.

Noise in the data

On the surface, it doesn't seem likely that noise could affect the performance of AI systems. After all, machines aren't affected by weather or football teams, so why would they make judgments that vary with circumstance? On the other hand, researchers know that bias affects AI, because it is reflected in the data that the AI is trained on.

For the new spate of AI models like ChatGPT, the gold standard is human performance on general intelligence problems such as common sense. ChatGPT and its peers are measured against human-labeled commonsense datasets.

Put simply, researchers and developers can ask the machine a commonsense question and compare it with human answers: “If I place a heavy rock on a paper table, will it collapse? Yes or No.” If there is high agreement between the two – in the best case, perfect agreement – the machine is approaching human-level common sense, according to the test.

Advertisement

So where would noise in? The commonsense question above seems simple, and most humans would likely agree on its answer, but there are many questions where there is more disagreement or uncertainty: “Is the sentence plausible or implausible? My dog plays volleyball.” In other words, there is potential for noise. It is not surprising that interesting commonsense questions would have some noise.

But the issue is that most AI tests don't account for this noise in experiments. Intuitively, questions generating human answers that tend to agree with one another should be weighted higher than if the answers diverge – in other words, where there is noise. Researchers still don't know whether or how to weigh AI's answers in that situation, but a first step is acknowledging that the problem exists.

Tracking down noise in the machine

Theory aside, the question still remains whether all of the above is hypothetical or if in real tests of common sense there is noise. The best way to prove or disprove the presence of noise is to take an existing test, remove the answers and get multiple people to independently label them, meaning answers. By measuring disagreement among humans, researchers can know just how much noise is in the test.

The details behind measuring this disagreement are complex, involving significant statistics and math. Besides, who is to say how common sense should be defined? How do you know the human judges are motivated enough to think through the question? These issues lie at the intersection of good experimental design and statistics. Robustness is key: One result, test or set of human labelers is unlikely to convince anyone. As a pragmatic matter, human labor is expensive. Perhaps for this reason, there haven't been any studies of possible noise in AI tests.

Advertisement

To address this gap, my colleagues and I designed such a study and published our findings in Nature Scientific Reports, showing that even in the domain of common sense, noise is inevitable. Because the setting in which judgments are elicited can matter, we did two kinds of studies. One type of study involved paid workers from Amazon Mechanical Turk, while the other study involved a smaller-scale labeling exercise in two labs at the University of Southern California and the Rensselaer Polytechnic Institute.

You can think of the former as a more realistic online setting, mirroring how many AI tests are actually labeled before being released for and evaluation. The latter is more of an extreme, guaranteeing high quality but at much smaller scales. The question we set out to answer was how inevitable is noise, and is it just a matter of quality control?

The results were sobering. In both settings, even on commonsense questions that might have been expected to elicit high – even universal – agreement, we found a nontrivial degree of noise. The noise was high enough that we inferred that between 4% and 10% of a system's performance could be attributed to noise.

To emphasize what this means, suppose I built an AI system that achieved 85% on a test, and you built an AI system that achieved 91%. Your system would seem to be a lot better than mine. But if there is noise in the human labels that were used to score the answers, then we're not sure anymore that the 6% improvement means much. For all we know, there may be no real improvement.

Advertisement

On AI leaderboards, where large language models like the one that powers ChatGPT are , performance differences between rival systems are far narrower, typically less than 1%. As we show in the paper, ordinary statistics do not really come to the rescue for disentangling the effects of noise from those of true performance improvements.

Noise audits

What is the way forward? Returning to Kahneman's book, he proposed the concept of a “noise audit” for quantifying and ultimately mitigating noise as much as possible. At the very least, AI researchers need to estimate what influence noise might be .

Auditing AI systems for bias is somewhat commonplace, so we believe that the concept of a noise audit should naturally follow. We hope that this study, as well as others like it, to their adoption.The Conversation

Mayank Kejriwal, Research Assistant Professor of Industrial & Systems Engineering, University of Southern California

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Advertisement

The Conversation

Vaccines tell a success story that Robert F. Kennedy Jr. and Trump forget – here are some key reminders

Published

on

theconversation.com – Mark R. O'Brian, Professor and Chair of Biochemistry, Jacobs School of Medicine and Biomedical Sciences, at Buffalo – 2024-07-26 07:11:29
Many fatal childhood illnesses can be prevented with vaccination.
Halfpoint Images/Moment via Getty Images

Mark R. O'Brian, University at Buffalo

Vaccinations have provided significant protection for the public against infectious diseases. However, there was a modest decrease in support in 2023 nationwide for vaccine requirements for children to attend .

In addition, the presidential candidacy of Robert F. Kennedy Jr., a leading critic of childhood vaccination, has given him a prominent platform in which to amplify his views. This includes an extensive interview on the “Joe Rogan Experience,” a podcast with over 14 million subscribers. Notably, former has said he is opposed to mandatory school COVID-19 vaccinations, and in a phone call Trump apparently wasn't aware was being recorded, he appeared to endorse Kennedy's views toward vaccines.

I am a biochemist and molecular biologist studying the roles microbes play in health and disease. I also teach medical and am interested in how the public understands science.

Advertisement

Here are some facts about vaccines that skeptics like Kennedy get wrong:

Vaccines are effective and safe

Public health data from 1974 to the present conclude that vaccines have saved at least 154 million lives worldwide over the past 50 years. Vaccines are also constantly monitored for safety in the U.S.

Nevertheless, the false claim that vaccines cause autism persists despite study after study of large populations throughout the world showing no causal link between them.

Claims about the dangers of vaccines often come from misrepresenting scientific research papers. Kennedy cites a 2005 report allegedly showing massive brain inflammation in monkeys in response to vaccination, when in fact the authors of that study state that there were no serious medical complications. A separate 2003 study that Kennedy claimed showed a 1,135% increase in autism in vaccinated versus unvaccinated children actually found no consistent significant association between vaccines and neurodevelopmental outcomes.

Advertisement

Kennedy also claims that a 2002 vaccine study included a control group of children 6 months of age and younger who were fed mercury-contaminated tuna sandwiches. This claim is false.

Gloved hands of clinician placing bandaid on child's arm, a syringe and vaccine vial beside them
Vaccines are continuously monitored for safety before and long after they're available to the general public.
Elena Zaretskaya/Moment via Getty Images

Aluminum adjuvants help boost immunity

Kennedy is co-counsel with a law firm that is suing the pharmaceutical company Merck based in part on the unfounded assertion that the aluminum in one of its vaccines causes neurological disease. Aluminum is added to many vaccines as an adjuvant to strengthen the body's immune response to the vaccine, thereby enhancing the body's defense against the targeted microbe.

The law firm's claim is based on a 2020 report showing that brain tissue from some with Alzheimer's disease, autism and multiple sclerosis have elevated levels of aluminum. The authors of that study do not assert that vaccines are the source of the aluminum, and vaccines are unlikely to be the culprit.

Notably, the brain samples analyzed in that study were from 47- to 105-year-old patients. Most people are exposed to aluminum primarily through their diets, and aluminum is eliminated from the body within days. Therefore, aluminum exposure from childhood vaccines is not expected to persist in those patients.

Vaccines undergo the same approval process as other drugs

Clinical trials for vaccines and other are blinded, randomized and placebo-controlled studies. For a vaccine trial, this means that participants are randomly divided into one group that receives the vaccine and a second group that receives a placebo saline solution. The researchers carrying out the study, and sometimes the participants, do not know who has received the vaccine or the placebo until the study has finished. This eliminates bias.

Advertisement

Results are published in the public domain. For example, vaccine trial data for COVID-19, human papilloma virus and rotavirus is available for anyone to access.

Vaccine manufacturers are liable for injury or death

Kennedy's against Merck contradicts his insistence that vaccine manufacturers are fully immune from litigation.

His claim is based on an incorrect interpretation of the National Vaccine Injury Compensation Program, or VICP. VICP is a no-fault federal program created to reduce frivolous lawsuits against vaccine manufacturers, which threaten to cause vaccine shortages and a resurgence of vaccine-preventable disease.

A person claiming injury from a vaccine can petition the U.S. Court of Federal Claims through the VICP for monetary compensation. If the VICP petition is denied, the claimant can then sue the vaccine manufacturer.

Advertisement
Gloved hand picking up vaccine vial among a tray of vaccine vials
Drug manufacturers are liable for any vaccine-related or injury.
Andreas Ren Photography Germany/Image Source via Getty Images

The majority of cases resolved under the VICP end in a negotiated settlement between parties without establishing that a vaccine was the cause of the claimed injury. Kennedy and his law firm have incorrectly used the payouts under the VICP to assert that vaccines are unsafe.

The VICP gets the vaccine manufacturer off the hook only if it has complied with all requirements of the Federal Food, Drug and Cosmetic Act and exercised due care. It does not protect the vaccine maker from claims of fraud or withholding information regarding the safety or efficacy of the vaccine during its development or after approval.

Good nutrition and sanitation are not substitutes for vaccination

Kennedy asserts that populations with adequate nutrition do not need vaccines to avoid infectious diseases. While it is clear that improvements in nutrition, sanitation, water treatment, food safety and public health measures have played important roles in reducing deaths and severe complications from infectious diseases, these factors do not eliminate the need for vaccines.

After World War II, the U.S. was a wealthy nation with substantial health-related infrastructure. Yet, Americans reported an average of 1 million cases per year of now-preventable infectious diseases.

Vaccines introduced or expanded in the 1950s and 1960s against diseases like diphtheria, pertussis, tetanus, measles, polio, mumps, rubella and Haemophilus influenza type B have resulted in the near or complete eradication of those diseases.

Advertisement

It's easy to forget why many infectious diseases are rarely encountered . The success of vaccines does not always tell its own story. It must be retold again and again to counter misinformation.The Conversation

Mark R. O'Brian, Professor and Chair of Biochemistry, Jacobs School of Medicine and Biomedical Sciences, University at Buffalo

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Read More

The post Vaccines tell a success story that Robert F. Kennedy Jr. and Trump forget – here are some key reminders appeared first on .com

Advertisement
Continue Reading

The Conversation

Tagging seals with sensors helps scientists track ocean currents and a changing climate

Published

on

theconversation.com – Lilian (Lily) Dove, Postdoctoral Fellow of Oceanography, Brown – 2024-07-25 07:08:14

Tagging seals with sensors helps scientists track ocean currents and a changing climate

Lilian Dove, Brown University

A surprising technique has helped scientists observe how Earth's oceans are changing, and it's not using specialized robots or artificial intelligence. It's tagging seals.

Several species of seals around and on Antarctica and regularly dive more than 100 meters in search of their next meal. These seals are experts at swimming through the vigorous ocean currents that make up the Southern Ocean. Their tolerance for deep waters and ability to navigate rough currents make these adventurous creatures the perfect research assistants to oceanographers like my colleagues and me study the Southern Ocean.

Advertisement

Seal sensors

Researchers have been attaching tags to the foreheads of seals for the past two decades to collect data in remote and inaccessible regions. A researcher tags the seal during mating season, when the marine mammal to shore to rest, and the tag remains attached to the seal for a year.

A researcher glues the tag to the seal's head – tagging seals does not affect their behavior. The tag detaches after the seal molts and sheds its fur for a new coat each year.

The tag collects data while the seal dives and transmits its location and the scientific data back to researchers via satellite when the seal surfaces for air.

First proposed in 2003, seal tagging has grown into an international collaboration with rigorous sensor accuracy standards and broad data sharing. Advances in satellite technology now allow scientists to have near-instant access to the data collected by a seal.

Advertisement

New scientific discoveries aided by seals

The tags attached to seals typically carry pressure, temperature and salinity sensors, all properties used to assess the ocean's rising temperatures and changing currents. The sensors also often contain chlorophyll fluorometers, which can data about the 's phytoplankton concentration.

Phytoplankton are tiny organisms that form the base of the oceanic food web. Their presence often means that animals such as fish and seals are around.

The seal sensors can also tell researchers about the effects of climate change around Antarctica. Approximately 150 tons of ice melts from Antarctica every year, contributing to global sea-level rise. This melting is driven by warm water carried to the ice shelves by oceanic currents.

With the data collected by seals, oceanographers have described some of the physical pathways this warm water travels to reach ice shelves and how currents transport the resulting melted ice away from glaciers.

Advertisement

Seals regularly dive under sea ice and near glacier ice shelves. These regions are challenging, and can even be dangerous, to sample with traditional oceanographic methods.

Across the open Southern Ocean, away from the Antarctic coast, seal data has also shed light on another pathway causing ocean warming. Excess heat from the atmosphere moves from the ocean surface, which is in contact with the atmosphere, down to the interior ocean in highly localized regions. In these , heat moves into the deep ocean, where it can't be dissipated out through the atmosphere.

The ocean stores most of the heat energy put into the atmosphere from human activity. So, understanding how this heat moves around helps researchers monitor oceans around the globe.

Seal behavior shaped by ocean physics

The seal data also provides marine biologists with information about the seals themselves. Scientists can determine where seals look for food. Some regions, called fronts, are hot spots for elephant seals to hunt for food.

Advertisement

In fronts, the ocean's circulation creates turbulence and mixes water in a way that brings nutrients up to the ocean's surface, where phytoplankton can use them. As a result, fronts can have phytoplankton blooms, which attract fish and seals.

Scientists use the tag data to see how seals are adapting to a changing climate and warming ocean. In the short term, seals may benefit from more ice melt around the Antarctic continent, as they tend to find more food in coastal areas with holes in the ice. Rising subsurface ocean temperatures, however, may change where their prey is and ultimately threaten seals' ability to thrive.

Seals have helped scientists understand and observe some of the most remote regions on Earth. On a changing planet, seal tag data will continue to provide observations of their ocean , which has vital implications for the rest of Earth's climate system.The Conversation

Lilian Dove, Postdoctoral Fellow of Oceanography, Brown University

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Advertisement

Read More

The post Tagging seals with sensors helps scientists track ocean currents and a changing climate appeared first on .com

Continue Reading

The Conversation

Cheesemaking is a complex science – a food chemist explains the process from milk to mozzarella

Published

on

theconversation.com – John A. Lucey, Professor of Food Science, of Wisconsin-Madison – 2024-07-24 07:18:57
Storing cheese wheels to let them age intensifies the flavor.
AP Photo/Antonio Calanni

John A. Lucey, University of Wisconsin-Madison

Cheese is a relatively simple food. It's made with milk, enzymes – these are proteins that can chop up other proteins – bacterial cultures and salt. Lots of complex chemistry goes into the cheesemaking process, which can determine whether the cheese turns out soft and gooey like mozzarella or hard and fragrant like Parmesan.

In fact, humans have been making cheese for about 10,000 years. Roman soldiers were given cheese as part of their rations. It is a nutritious food that provides protein, calcium and other minerals. Its long shelf allows it to be transported, traded and shipped long distances.

I am a food scientist at the University of Wisconsin who has studied cheese chemistry for the past 35 years.

Advertisement

In the U.S., cheese is predominantly made with cow's milk. But you can also find cheese made with milk from other animals like sheep, goats and even buffalo and yak.

Unlike with yogurt, another fermented dairy product, cheesemakers whey – which is water – to make cheese. Milk is about 90% water, whereas a cheese like cheddar is less than about 38% water.

Removing water from milk to make cheese results in a harder, firmer product with a longer shelf life, since milk is very perishable and spoils quickly. Before the invention of refrigeration, milk would quickly sour. Making cheese was a way to preserve the nutrients in milk so you could eat it weeks or months in the future.

How is cheese made?

All cheesemakers first pump milk into a cheese vat and add a special enzyme called rennet. This enzyme destabilizes the proteins in the milk – the proteins then aggregate together and make a gel. The cheesemaker is essentially turning milk from a liquid into a gel.

Advertisement

After anywhere from 10 minutes to an hour, depending on the type of cheese, the cheesemaker cuts this gel, typically into cubes. Cutting the gel helps some of the whey, or water, separate from the cheese curd, which is made of aggregated milk and looks like a yogurt gel. Cutting the gel into cubes lets some water escape from the newly cut surfaces through small pores, or openings, in the gel.

The cheesemaker's goal is to remove as much whey and moisture from the curd as they need to for their specific recipe. To do so, the cheesemaker might stir or heat up the curd, which helps release whey and moisture. Depending on the type of cheese made, the cheesemaker will drain the whey and water from the vat, leaving behind the cheese curds.

A man in a white lab coat, hairnet and gloves pulls a device through a large tub of white liquid.
Wisconsin Master Cheesemaker Gary Grossen cuts a vat of cheese with a cheese harp during a cheesemaking short course at the Center for Dairy Research in Madison, Wis. Cutting helps release whey during the cheesemaking .
UW Center for Dairy Research

For a harder cheese like cheddar, the cheesemaker adds salt directly to the curds while they're still in the vat. Salting the curds expels more whey and moisture. The cheesemaker then packs the curds together in forms or hoops – these are containers that shape the curds into a block or wheel and hold them there – and places them under pressure. The pressure squeezes the curds in these hoops, and they knit together to form a solid block of cheese.

Cheesemakers salt other cheeses, like mozzarella, by placing them in a salt solution called a brine. The cheese block or wheel floats in a brine tank for hours, days or even weeks. During that time, the cheese absorbs some of the salt, which adds flavor and protects against unwanted bacterial or pathogen growth.

A graphic showing the many steps between a farmer harvesting milk from cows and the cheese reaching the consumer.
The cheese production process.
UW Center for Dairy Research

Cheese is a living, fermented food

While the cheesemaker is completing all these steps, several important bacterial processes are occurring. The cheesemaker adds cheese cultures, which are bacteria they choose that produce specific flavors, at the beginning of the process. Adding them to the milk while it is still liquid gives the bacteria time to ferment the lactose in the milk.

Historically, cheesemakers used raw milk, and the bacteria in the raw milk soured the cheese. Now, cheesemakers use pasteurization, a mild heat treatment that destroys any pathogens present in the raw milk. But using this treatment means the cheesemakers need to add back in some bacteria called starters – these “start” the fermentation process.

Advertisement

Pasteurization provides a more controlled process for the cheesemaker, as they can select specific bacteria to add, rather than whatever is present in the raw milk.
Essentially, these bacteria eat (ferment) the sugar – the lactose – and in doing so produce lactic acid, as well as other desirable flavor compounds in the cheese like diacetyl, which smells like hot buttered popcorn.

In some types of cheese, these cultures stay active in the cheese long after it leaves the cheese vat. Many cheesemakers age their cheeses for weeks, months or even years to give the fermentation process more time to develop the desired flavors. Aged cheeses include Parmesan, aged cheddars and Gouda.

A person in a white coat holds a wheel of cheese.
A Wisconsin cheesemaker inspects a wheel of Parmesan in the aging room. Aging is an important step in the production of many cheeses, as it allows for flavor .
The Dairy Farmers of Wisconsin

In essence, cheesemaking is a milk concentration process. Cheesemakers want their final product to have the milk proteins, fat and nutrients, without as much of the water. For example, the main milk protein that is captured in the cheesemaking process is casein. Milk might contain about 2.5% casein content, but a finished cheese like cheddar may contain about 25% casein (protein). So cheese contains lots of nutrients protein, calcium and fat.

Infinite possibilities with cheese

There are hundreds of different varieties of cow's milk cheese made across the globe, and they all start with milk. All of these different varieties are produced by adjusting the cheesemaking process.

For some cheeses, like Limburger, the cheesemaker rubs a smear – a solution containing various types of bacteria – on the cheese's surface during the aging process. For others, like Camembert, the cheesemaker places the cheese in an (e.g., a cave) that encourages mold growth.

Advertisement

Others like bandaged cheddar are wrapped with bandages or covered with ash. Adding a bandage or ash onto the cheese's surface helps protect it from excessive mold growth, and it reduces the amount of moisture lost to evaporation. This creates a harder cheese with stronger flavors.

A man in a white apron and hat stands in a room full of shelves stacked with cheese.
Wisconsin Master Cheesemaker Joe Widmer in his brick cheese aging room. Brick cheese is a smear-ripened cheese – it is produced by applying a salt solution to the exterior of the cheese as it ages.
Dairy Farmers of Wisconsin

Over the past 60 years, cheesemakers have figured out how to select the right bacterial cultures to make cheese with specific flavors and textures. The possibilities are endless, and there's no limit to the cheesemaker's imagination.The Conversation

John A. Lucey, Professor of Food Science, University of Wisconsin-Madison

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Read More

Advertisement

The post Cheesemaking is a complex science – a food chemist explains the process from milk to mozzarella appeared first on .com

Continue Reading

Trending