How ChatGPT Recipes Compare to Chef Recipes

How ChatGPT Recipes Compare to Chef Recipes

The heat generated by ChatGPT and other artificial intelligence software has entered the kitchen — or at least the kitchens of TikTok, where creator Michelle Meng, aka @hashslingers, has been pitting recipes generated by AI against those created by professional chefs, restaurant chains, food brands, and the occasional celebrity. With over a dozen videos released so far, Meng’s ongoing experiment has included McDonald’s apple pie, Claire Saffitz’s focaccia bread, and Taylor Swift’s chai sugar cookies, each of which Meng has pitted against an AI-generated recipe for the same dish.

The project, which was inspired by Patrick Zeinali’s YouTube short “Robot Vs Human Bake Off,” was born earlier this year after Meng was laid off from her job as a software engineer. Meng started @hashslingers in 2021 as a creative outlet — she also has a YouTube channel — and, following her layoff, has decided to give herself a year to pursue her passions. We talked to her about the winning and losing recipes, ChatGPT as a cooking tool, and the debate around AI.

Eater: What have you found so far? Do real chefs tend to win out in this, or does AI?

Michelle Meng: It’s really been a toss-up. Mostly, chefs and restaurants and creators typically win, but there hasn’t been an obvious pattern.

I’ve noticed that with dishes that require very technical skills, like baking or using yeast, the AI does okay, as in it’s edible. But the chefs usually take the wins on those because the AI’s directions aren’t really fleshed out. Baking’s just very exact. In cultural dishes, like Uncle Roger’s fried rice or Golden Gully’s mango kulfi, the chef typically wins because they better understand the background and the flavor of the cuisine.

[But] AI does surprisingly well every time. It wins mainly against food chains, and occasionally other recipes, because it throws in more complex ingredients.

What recipes have you noticed that AI does really well?

Mainly food chains. We’ve done Panda Express orange chicken and In-N-Out animal style fries.

I think some of it has to do with actually cooking it at home, and giving attention to the food, versus just assembling the food at a fast food chain. I do try to keep that in mind when doing the judging.

The only case where the AI has actually won against the chef is with Gordon Ramsay’s scrambled eggs, which was really surprising.

What do you think this means for recipe creation?

There’s been a lot of debate about this in my comments. Similar to the conversation about AI-generated art, there’s the notion that AI is taking chefs’ and food bloggers’ online recipes — which are their creative and intellectual property — and using them without due credit. There is a discussion to be had there.

In general, with the rise of AI and machine learning, a lot of people are scared that it’s taking over. But as someone who’s studied and built this technology, it’s difficult to reassure people that it’s more of a tool and less of a replacement.

In my opinion, things like ChatGPT are a great cooking tool. If I was a new chef or a home cook who wants to learn more, ChatGPT is a single source that I can go to ask for a recipe and subsequently ask for different cooking tips or techniques. It’s less effort than scouring the internet for different recipes. But in terms of recipe creation, it’s hard to replicate the history, emotion, and human touch that’s so important in food.

Do you think then that we should be worried about AI in the recipe space?

I’m not sure if I stand firmly on one side or the other. AI is kind of doing what humans do, but on a larger and faster scale. When humans watch content or look up recipes or even look at art, we consume that knowledge, and it impacts what we do. If I’m researching recipes, I note the directions and ingredients, and I make it my own. But it is inspired by other content that I’ve consumed before. Technically speaking, I don’t necessarily think AI is violating intellectual property, but, morally, I’m unsure.

Do you think that using AI in cooking is going to grow more popular? Have you seen other creators follow your lead?

I’ve seen a couple of similar videos pop up on my TikTok. I find it super interesting to see people trying different chefs, recipes, and techniques. It has been a pretty popular thing on social media right now, and I hope people actually give it a try.

I’m not sure if it’s just a fad of the times. Right now, ChatGPT and AI is a pretty popular topic, so I hope that it stays relevant and people find it a useful tool.

Have you noticed whether the ChatGPT recipes are getting better as time goes on?

I’ve noticed that all of the recipes actually work out pretty well, and sometimes they’re super tasty. I’ve never had one fail horribly. As the videos go on, I see that the directions get more sophisticated in terms of describing the actual steps of the process, which would be really helpful to people who are going to use ChatGPT as a cooking tool.

What do you want to see come out of the experiment? Is it a fun thing to do, or do you hope that it makes people think differently about how AI can be used in the kitchen?

A little bit of both. It was just a fun way to merge tech and cooking, which are my two passions.

The initial thought behind the experiment was whether AI could be a useful tool in the kitchen. I’d like to make people more open to the idea of trying out new technology, especially with cooking, since it’s an area where tradition is a big part of it, and things are usually done a very specific and certain way.

Is your plan to continue doing this indefinitely, and see how AI evolves with cooking? Or is there another goal?

I don’t have anything seriously planned out right now, but I would love to continue the series if people are enjoying it and getting something from it — and, somehow, maybe evolve it.

Ultimately, do you foresee AI replacing humans in the kitchen?

I don’t think that AI will be replacing humans in the kitchen. It’s just an interesting tool to try. And if it’s useful, that’s cool.

This interview has been edited and condensed for clarity.

Megan Ulu-Lani Boyanton reports on the business beat at the Denver Post, and previously covered agriculture and trade policy for Bloomberg Government. She also writes for Eater, Smithsonian Magazine, Delish and more.

In the US, the definition of “healthy food” is changing – how does this compare to Australia?

In the US, the definition of “healthy food” is changing – how does this compare to Australia?

In the US, the definition of healthy food stuff is switching – how does this examine to Australia?

In September 2022, the United States Foods and Drug Administration (Food and drug administration) proposed updating the nutritional requirements that apply when a foodstuff item needs to declare to be “healthy”, with the said goal of encouraging US shoppers to eat healthier food and – by extension – lower incidence of persistent ailment. This post considers how that proposed requirements aligns or differs with Australian polices as to when a foodstuff product can claim to be “healthy”.

What are the proposed U.S. alterations?

Nutritional science has occur a extensive way considering that 1994, when the Food and drug administration first launched the dietary conditions for “healthy” foodstuff. The present proposal is an update of current restrictions which previously imposed restrictions on fats, saturated body fat, cholesterol and sodium as well as minimal portions of Vitamin A, Vitamin C, calcium, iron, protein and fibre before a meals was qualified to claim to be “healthy”.

Why the modify? Part of the dilemma with the current tactic is that – by implementing a total of diet paradigm to unique foods – quite a few food items which dietary science needs to motivate have been not eligible to assert to be “healthy”. For instance, some total raw fruits and veggies did not fulfill the nutritional requirements and for that reason could not be described as “healthy”. On top of that, the focus on complete fats ended up punishing some products that have been the natural way substantial in “healthier” unsaturated fatty acids.

As a result, the most sizeable change set forward by the proposal is a entire of eating plan solution (“emphasise healthful nutritional patterns”) instead than focusing on particular person vitamins.

The new approach needs a “healthy” foodstuff to include a particular amount of substances from a proposed food team (specifically vegetables, fruits, grains but at minimum 50 percent wholegrain, dairy, protein and oils) in addition to minimum nutritional profiles. In addition, additional sugars are extra as a damaging pounds on the dietary standards.

As an addition to altering the criteria, the Food and drug administration has introduced that it is checking out introducing a entrance of pack visible representation of the “healthy” claim.

How does this evaluate with Australia?

The very best position of comparison in Australian is the Nutrient Profiling Scoring Criterion (NPSC), introduced into the Australia New Zealand Food items Benchmarks Code as section of Typical 1.2.7 in 2013. The NPSC is the criterion that can use (but not generally) when a foodstuff products wishes to make a “health claim”.

In some techniques, the new proposed Fda criteria is really similar to the NPSC. The NPSC is a go/fall short binary and, in examining any individual product in opposition to the NPSC, it punishes food items products for their power, saturated body fat, sugar and sodium information and benefits products for their protein, fibre and fruit and vegetable information.

Like the new Fda method, the NPSC does not punish whole extra fat and as an alternative focuses on saturated extra fat (but leaves trans fats on the table). By way of distinction, the NPSC punishes total sugar content, instead than focusing on added sugar.

Nonetheless, the largest discrepancies the new Food and drug administration tactic requires when in contrast to the NPSC are that the Food and drug administration approach:

  • sets minimum amount meals team information and consequently arguably encourages consumption of nutrient dense meals in a way that the NPSC does not and
  • sets its dietary criteria in different ways based on the food team from which the item comes.

The Fda technique demonstrates existing dietary science in a way that the NPSC fails to do. The NPSC sets its nutritional criteria for only 3 distinct groups of product or service: beverages, oil-centered products this kind of as cheese, butter, and so on, and then every thing else. This usually means that the very same dietary conditions utilize to uncooked greens as to muesli bars or chocolate or potato chips or pasta or … I could go on. This deficiency of nuance in making use of the NPSC to unique food groups signifies that some evidently unhealthy foods go the criterion though some wholesome food items do not, comparable to the place the prior US framework uncovered by itself in.

In other text, the NPSC proceeds to broadly utilize a full of eating plan paradigm to individual food items, relatively than encouraging much better dietary patterns. By way of further contrast, the voluntary overall health star rating program (Australia’s very own front-of-pack graphic to inspire healthier consuming) incorporates far a lot more meals types and gradations in scale than the legislated NPSC.

The Australian authorities would do very well to adhere to the current method being regarded in the US.

Distinguishing in between “healthy” claims and “health” promises in Australia

It is really worth noting that, in Australia, the NPSC only applies when generating a “health claim”, and not necessarily when proclaiming that a food item is “healthy”. When it would make sense to contemplate a “healthy” declare to be inherently a “health claim”, the authorized definition of a “health claim” is a whole lot extra confined than its identify suggests.

For the NPSC to apply, a food product or service have to indicate that it has a “health effect”. A “health effect” is outlined as:

an outcome on the human physique, such as an outcome on a single or far more of the next:

(a) a biochemical procedure or end result

(b) a physiological system or final result

(c) a functional system or result

(d) advancement and enhancement

(e) actual physical general performance

(f) psychological functionality

(g) a disease, ailment or problem.

Does the declare “healthy” by by itself in fact imply a well being impact on the human overall body? If so… what is it? What is the bodily functionality or biochemical outcome of a consuming a “healthy” meals? What does the phrase “healthy” necessarily mean in and of itself?

The restrict of the legal definition of a “health claim” was explored in the New Zealand Court of Appeal case in 2016 between the New Zealand Ministry of Major Industries (the NZMPI) and the proprietor of the “Manuka Doctor” honey brand. In that scenario, the NZMPI was refusing to grant an export licence due to th “Manuka Doctor” trademark becoming a non-compliant wellness declare on behalf of the item. However, the Court of Attraction discovered that even though the trademark “Manuka Doctor” implied “healthiness”, it did not suggest a exact health and fitness outcome.

As a consequence, it could be argued that in Australia no approved requirements utilize to food items products and solutions that assert to be “healthy” devoid of implying a exact wellness profit. Nevertheless, purchaser protection legal guidelines also apply, and hence solutions will have to not represent “healthy” (or other) statements which are deceptive or deceptive in breach of people legal guidelines (as famously transpired when the ACCC correctly prosecuted Heinz Shredz for deceptive customers from 2006 to 2018 that a higher sugar solution was wholesome).

Hence, any food products wishing to assert to be “healthy” in Australia devoid of meeting the NPSC ought to:

(a) steer clear of any other claims which could imply a exact overall health impact and

(b) have some substantiation proof as to how they are “healthy”.

It is worthy of noting that when Grill’d burgers was regularly assessed by AdStandards in 2021 for such as “healthy” in their tag line, 1 of their evidence details that the claim was not misleading was that several of their items basically fulfilled the NPSC.