New Flash: AI Won't Fix Obesity

 

Dr Emrick's Books and Articles

 

Obesity is not a failure of character; it is a predictable outcome when cheap, hyper-palatable food meets biology sculpted for scarcity. AI will make care more precise and more equitable in many areas of medicine. It will not lower the calorie density of takeout or slow the speed at which we eat. The only plausible path to a different map with more yellow and green is a broad shift in eating habits, reinforced by environments that make the better choice the easy choice. That shift happens one plate at a time, one family at a time, and, if leaders are brave, one individual discussion at a time.

Look closely at the CDC’s 2023 obesity map, Adult Obesity Prevalence Maps | Obesity | CDC, and you see a near-solid field of orange and red. In every state and U.S. territory, at least one in five adults lives with obesity, and in three states, the share exceeds 40 percent. The burden tilts toward the South and Midwest, and the pattern mirrors differences by education and race or ethnicity. These are not static features of the landscape; they translate into clinical volumes, chronic illness, surgical risk, disability, and dollars. Public health dashboards often reduce this reality to a color scale. Hospital leaders and payers feel it in budgets, staffing, and the slower, grinding work of managing preventable disease. Two lines of evidence sharpen the cost picture.

First, careful econometric work using instrumental variables, a method that aims to estimate causal effects, finds that obesity drives significant increases in direct medical spending nationally and in the most populous states. Using recent surveys and expenditure data, researchers estimated the portion of expenditures that obesity causes rather than merely correlates with. Second, recent analyses translate weight loss into spending changes. Using U.S. claims and survey data for people with employer insurance and Medicare beneficiaries, a 2024 cross-sectional study projected that losing five percent of body weight is associated with roughly eight percent lower annual health care spending for working-age adults and about seven percent lower spending for older adults with comorbidities, with larger savings at 10–25 percent weight loss. For Medicare enrollees with obesity, each one-point rise in BMI added an estimated $633 in yearly costs. These are not modeled fantasies; they reflect patterns in real spending that repeat across datasets.

With costs rising and new anti-obesity medications on the market, it is tempting to believe that technology, especially AI, will solve the problem. AI can support triage, personalize nudges, or predict risk. Meta-analyses show that digital programs and app-based coaching can produce small to moderate short-term weight losses and can match some traditional counseling programs under trial conditions. Yet these effects tend to be modest, hinge on sustained engagement, and fade when people stop using the tools. AI can personalize advice; it cannot swallow our food for us. The same hard limit appears with the much-discussed GLP-1 and dual-agonist drugs. Randomized trials show impressive average weight loss while on therapy. But when treatment stops, people regain a large share of the weight they lost, and cardiometabolic benefits drift back toward baseline. Economists also point out that, at current U.S. prices, newer agents are not cost-effective for broad coverage without huge discounts; extending public coverage would carry a substantial budget impact even before accounting for long-term adherence. Medications can be valuable for selected patients, particularly with severe obesity or weight-related disease, yet they do not remove the need to change how and what we eat.

A common objection goes like this: “We have argued about low-fat and low-carb for decades. If experts disagree, how can individuals act?” The best trials resolve the false choice. When researchers compared well-constructed low-fat and low-carbohydrate diets and emphasized whole foods, both groups lost similar amounts of weight over one year; genotype patterns and baseline insulin responses did not predict who did better. The practical message is clear: pick a pattern you can sustain that prioritizes minimally processed foods, adequate protein, fiber, and thoughtful portions. Experiments also show why modern eating environments sabotage intentions. In a tightly controlled inpatient crossover trial, adults randomized to an ultra-processed diet consumed about 500 more calories per day and gained weight within two weeks, even though meals matched macronutrients, sugar, sodium, and fiber. Energy density and speed of eating are likely to explain much of the effect. Portion size amplifies the problem: laboratory and field studies show that larger portions and higher energy density drive greater intake, often outside conscious awareness. These levers sit on our plates at every meal; no algorithm can countermand them if the food is already in our hands.

Skeptics sometimes argue that people cannot maintain lifestyle changes. The Diabetes Prevention Program and its long-term follow-up offer a more comprehensive and robust story. Intensive changes in eating and physical activity cut diabetes incidence by more than half in the original trial and preserved meaningful benefits years later. Weight loss averaged only modest amounts, yet the risk reduction persisted. The point is not to chase perfection; it is to reinforce daily patterns that inch risk downward and keep costs from compounding. Hospitals, payers, and employers often seek a lever to move populations quickly. AI will help us find high-risk patients, target outreach, and monitor progress at scale. It can sort through food logs, predict who is about to lapse, and cue a text at the right moment. Those are useful edges. They do not change the law of conservation of energy. Without a change in the average American plate, the map will not budge.

So, what are solid programs to help support a healthier community? First, support programs that train people to shop, cook, and portion real food. The Medicare Diabetes Prevention Program exists for a reason and has strong evidence. Tie benefits to sustained participation and measurable dietary quality, not just steps or app logins. Use pharmacy-and-therapy coverage rules that prioritize value: GLP-1 coverage when clinically indicated, with clear deprescribing and lifestyle maintenance plans, and price negotiations tied to cost-effectiveness thresholds. Employers can reshape cafeterias and vending contracts, subsidize healthier options, and end the default of ultra-processed snacks at meetings. These choices influence intake more than any chatbot ever will. Second, healthcare providers must speak plainly and openly about food. Focus on patterns over macros and keep the counsel concrete: eat mostly food with one ingredient; build meals around protein and plants; curb ultra-processed items you would not recognize in a home kitchen; watch portions, primarily beverages and energy-dense snacks. Explain that sustainable change beats short bursts of austerity. When you prescribe GLP-1s, attach a food plan and a taper strategy to reduce rebound risk if the medication ends. The weight-regain data make that step more than a courtesy; it is clinical prudence. GPL-1s are not a good and viable long-term solution. The long-term solution is better eating habits day in and day out. For individuals, begin where you can win. Replace one ultra-processed breakfast with eggs, yogurt, oats, or fruit and nuts. Eat the same simple lunch on workdays to remove fatigue. Cook in batches so dinner needs less willpower. Serve meals on smaller plates to exploit the portion-size effect in your favor and pour sweet beverages into glasses you cannot refill automatically. If you use an app or AI assistant, treat it like cruise control on a long drive: helpful on straight roads, useless without your hands on the wheel.