Hacker News Reader

Comment by karaterobot | Hacker News Reader

> The new CEO, she said, “thinks that the [technology] can do better than us, and that clients don’t care ... that there’s not a person behind the computer.”

That's surprising. The only reason I've ever considered Stitch Fix is specifically because there might be a person with better taste than me on the other end.

An algorithm is just going to give me a mix of what's popular and what I've liked in the past, which is precisely the information I already had before coming to their service. Wouldn't people eventually realize they don't need to be told to keep buying things they already like?

I'd be more interested if they not only kept the people around, but doubled down on having consistent relationships and interaction between stylists and clients.

View on HN

The only reason I've ever considered Stitch Fix is specifically because there might be a person with better taste than me on the other end.

If the ex-workers in the article are representative, they aren't trained stylists at all; they're housewives and other normal people looking for extra income. Using StitchFix is basically letting your Uber driver select your wardrobe.

If they have good taste, works for me.

I'd venture a guess that the average "normal person" is better at picking out clothing than I am, so I don't see the problem with this. Vanishingly few people who choose clothes for others, whether it's a Stitch Fix, Nordstrom, or styling celebrities, are "trained stylists." It's about taste, not training.

> An algorithm is just going to give me a mix of what's popular and what I've liked in the past.

That's exactly the same thing a human stylist would do + making sure the outfit pieces look good together (Which I assume the algorithm is also trained to do).

As much as I want to believe that a human in the end would do a better job, I think an algorithm is capable of becoming a more accurate and dynamic stylist than a human.

Unless you want to be a trend-setter or do some artistic expression through your clothing. In those cases a human stylist does make sense. But for the regular Joe, I think a well trained algorithm can perform better than a human.

This is just a misunderstanding of how creative people work vs algorithms.

An algorithm will aim to make you look similar to what other people like you look like. It will not push boundaries or riff on creative 'happy accidents' because it can't.

A real stylist will do those things and more.

I don't know whether it's an age thing or just pure cognitive dissonance because people in tech have the hubris to think we can optimise and improve everything because technology, but this machine-learning nihilist thinking is profoundly sad.

A real stylist can do that, but very well may not. The question is not whether an algorithm/AI can beat stylists at the 90th percentile of how great they are, it's whether it can beat stylists at the 50th percentile.

Sorry but could you please give a definition of creativity which is not a weaker or more limited version of mine:

> Creativity is the preference for low(er) priority edges in a graph

There is virtually no difference between an algorithm and a human except that the human may have better heuristics for what "works", which by definition is a form of bias and an algorithm can assess many orders of magnitude more nodes.

If Lee-Sedol the world champion at Go could be left dumbfounded by what he claimed to be a creative move, a move generated by a laptop performing monte-carlo tree search with an ANN doing pruning, and have his world shattered, then I don't see how anyone else could claim that algorithms can't be creative.

If we go by something like this [1]:

> Creativity is defined as the tendency to generate or recognize ideas, alternatives, or possibilities that may be useful in solving problems, communicating with others, and entertaining ourselves and others. (page 396)

Then the tendency is reweighting of priorities of heuristics, the generation of ideas is walking the varius edges, recognizing is nothing but an evaluation of the new state, and alternatives and posibilities that may be useful is nothing but searching deeper within the graph.

[1] https://www.csun.edu/~vcpsy00h/creativity/define.htm

I would appreciate any comments that explain what is wrong with my definition or the comment as a whole. I do not find this downvoting without any form of discussion to be of substance, nor facilitate any discussion, it achieves nothing except leave the author perplexed but without a way of reconciling it, except by assuming that people dislike hearing what they don't want to hear, that is, that human brains are nothing, but heuristic based computational units with lazy evaluation.

Grand parent comment claimed that believing algorithms (and computers) can perform better than humans is hubris, I claim that believing humans are special, is hubris.

I don't disagree with what you're saying in a general sense. Creative people output will be better and higher quality than an algorithm...But I don't think that's the case for this business model.

Although I believe there is creative merit in composing an clothing outfit I don't believe that merit is even remotely close to things such as designing the actual clothes or creating the drawings and art that goes into the clothes or designing the stores in which these clothes are sold...

So I stand by my original statement. I do believe an algorithmic stylist for the avergage person can be vastly superior than a human stylist, because the inputs that the former takes are not that different than the inputs of the latter. Furthermore, the reasoning is essentially the same:

People want good fitted clothes, that are in fashion and that go in accordance to what they like. I don't think there's a lot of room for creativity here. I'm sorry that it sounds like a machine-learning nihilist take, but in practice I don't believe there's a good match between what people want from something like Stitch Fix and what a human stylist can do for them.

I dooubt that someone who relates to clothing in an artistic way would ever use something like Stitch Fix, so my assumption is that for the average Stitch Fix user, an algorithmic stylist could give them the same desired output at a better cost.

> An algorithm will aim to make you look similar to what other people like you look like. It will not push boundaries or riff on creative 'happy accidents' because it can't.

An algorithm can aim to push boundaries and riff on creative 'happy accidents' in ways that people like you like. It's not even that explicit with typical supervised learning/recommenders. It will find things people like you like, whether that means pushing boundaries in ways those people like or just making you look like they like.

I feel the same as you do about ML being in too many places too confidently, but you're assuming a lot about the kinds of algorithms used.

There is no way for you to buck the pattern here since it is using a statistical estimation for whether you like something. You could very well be an outlier datapoint rather than someone in the median, at which point the model fit for you is extremely poor. If you had an actual person on the other end you could tell them explicitly what you want rather than have some likelihood estimation based on metadata that might not have any relevance to this particular decision you are making at this time.

I think you missed this part of their comment:

> Unless you want to be a trend-setter or do some artistic expression through your clothing. In those cases a human stylist does make sense. But for the regular Joe, I think a well trained algorithm can perform better than a human.

And your comment:

> An algorithm will aim to make you look similar to what other people like you look like. It will not push boundaries or riff on creative 'happy accidents' because it can't.

Correct - I think that's what most people want. Style is defined at the high end to a very limited audience and derivative variations of it are built on further derivative variations as it flows down the classes until it so distorted it simply dissolves. Every turn of that crank is another group of people trying to look like other people (and thus offending the tastes of the group being copied).

> Style is defined at the high end to a very limited audience and derivative variations of it are built on further derivative variations as it flows down the classes

That is a very clichéd haute couture definition — a la “Devil Wears Prada - Cerulean Top” https://www.youtube.com/watch?v=vL-KQij0I8I

That trickle-down definition of style is relevant perhaps within some groups of people.

I think that most subcultures have their own highly independent styles, and within that you have people with their own independent riffs on the general subculture style.

Do you not think that many subcultures have their own style that isn’t derivative/trickle-down?

For sure. Certain subcultures have aesthetics that are not started at the top. However, these subcultures at times could be thought of as culturally elite (and very much adored by upper class tastemakers) and their fashion in some cases manages the same derivative cycle. For instance, “Grunge” culture which had an aesthetic that was very much not haute of course. But it was culturally elite at the time and very quickly their styles were derived for mainstream consumption. But even so, haute will deconstruct the parts they like, integrate it, and so it passes.

Other subcultures of course never break through.

Regardless, most clothing is purchased from retailers that mass produce things. You could probably tier rank them accordingly from designers to boutiques to small retailers on up to big box shops that you can find anywhere.

You see the same thing in other cultural artifacts like cuisine. Haute Cuisine doesn’t start every trend but it isn’t afraid to discover something outside the mainstream, riff on it, and then it works it’s way into the middle.

It's honestly the latter.

The algo literally gives 10 cargo jackets in the summer because they are “high match scores”.

The algorithm sends winter coats and three crop tops to someone living in Texas during the summer.. it doesn’t put together outfits, consider seasons, or make selections based on budget/requests/lifestyles/etc.

So a human (usually with a fashion background and retail experience) is going to be able to perform better in pretty much every scenario with their algorithm “helping”

I use Stitch Fix and personally I don't care. So far the selections have been great.

It sounds like the stylist is provided data from an algorithm already any way and just fine tunes the selections.

Not only that they have a lot of ways to build out the style you already received which is great.

I mean the whole gimmick is about off-loading hard-to-sell inventory at immense mark-ups by bundling items together. Yea some absurd combinations will be sent back but it will still profit an order of magnitude more then displaying it in a department store for months. A lot more room for JIT style and data-driven logistics without the lag-time of presenting within a store.

You can't be that mad at a company when it is so obvious what their value proposition and consequences are to a consumer.

You can see the difference in recommendation by something like Trunk Club which has the entire Nordstrom catalog to pull from. The stylists who work there also tend to be higher quality.

The difficulty with JIT/data-driven logistics in this context is that Stitch Fix's styled offering (the "Fix" boxes) all has to be fulfilled from a single distribution center. So you need not only to predict demand for particular articles of clothing but to ensure that there's sufficient inventory of items to go along with them. It's easy to get people to buy one piece and return four, but does that sound more like a "personal styling service" or a subscription box?

I'm still wondering how many "stylists" are good enough for the job. Or maybe I'm just too picky and thus find it hard to imagine that many stilists would be able to find what the other person would like without it being too obvious (hey, I noticed you liked blue things. here's another blue one!). In my imagination it's a very thin line between obviousness and misunderstanding (and the fine line being the absolute understanding of the customer).

Yes, it's very difficult to tell your style from your photos, social media presence, and explicit instructions provided when you signed up. This is sarcasm if you didn't pick it up.

People who are as picky as you outlined self-select out of services like this.

What they have that the customer doesn't, is knowledge of the space of products that may match your taste. They also know more brands than you do, and may match you with one that you weren't aware of. Like most services, we can do it ourselves, but we pay others because we don't have the time.

yeah, you're probably right. It should get easier to find adequate stylings for people who stay longer with the service, as the database fills. And that's probably also the moment where less stylists are needed because your "AI" can do (more of) it.

btw. how they treated their workers judging from the article is absolute appaling and another indictment of gig companies.

You also do not want just what "I like". A stylish friend did a shopping trip with me and made me buy stuff I didn't really like: too tight, too colourful, etc. Not outlandish, just way outside my comfort zone. Apparently I look really good in red.

OMG. The compliments. From co-workers, random people on the street and staff in stores. I was shopping for a nice watch for my wife, went to a place and was ignored but saw a watch, went back a few days later in my new clothes, and the same staff was super friendly and helpful. It really changed how I viewed the importance of style and fit.

It's not like you are getting top-tier fashion designers, interior decorators etc. on services like these. The company has a fixed catalog, decided based on what deals they can get from vendors, and their reps are all outsourced and following a script/reading out algorithmic results.

"In the future, your job will be performed by a robot, but the robot will be you."

Im the same. I don't mine the algo but i wanted a human as well. This is disappointing. Ive been really pleased with what they sent me.

Would HN recommend Stitch Fix? I was thinking of signing up if there was a personal stylist recommending clothes, but now I’m not sure. How good is the AI in picking a wardrobe?

How does a stylist learn “style” and why could an algorithm not perform the same?

Reply on news.ycombinator.com