Does CGC use computer vision or AI at any stage of the encapsulation lifecycle? 1 1

[707comics]

I was pondering the QA threads I have been reading lately and I got to thinking. It seems with the "checks & balances" that are in place during the grading process there would be a very low chance that 2 or 3 humans looking at the book during the acceptance, grading, QA and shipping process would not catch errors like wrong book being encapsulated or mismatched titles / issues on label (not even talking about under / over grading issues). 

From my experience, systems that are used to enhance throughput of sorting and routing tend to use some type of computer vision and/or AI to identify, route and track things throughout the product lifecycle at the warehouse. I was wondering if anything like that is being used (at any point) of the books stay at CGC (acceptance / sorting, grading, encapsulation, shipping, etc.) . Is anybody aware of any type of this technology being used by CGC (or competitors)? 

[adampasz]

I doubt it.

There are potentially interesting applications for these technologies, but I suspect the cost to implement them into CGC's workflows would be high.

A lot of AI/ML solutions look impressive, but they only get you about 80% accuracy at best. Unless you are implementing the solution at very large scale, it is difficult for a business to justify the investment.  Even if you had an AI that graded comics correctly 80% of the time, you'd still need humans overseeing everything. If you have 1000 full-time graders, you'd still need to retain at least 200 of them, plus management, training, and other overhead. Compare the cost to what you would pay to develop and maintain the technology.  It makes the most business sense if you can get rid of all your human graders, or if your business is big, it is no longer feasible to hire more graders.

Maybe this will change as the tech improves, but I suspect that's still a decade away. 

 

Edited October 1, 2022 by adampasz

[707comics]

On 10/1/2022 at 11:15 AM, adampasz said:

I doubt it.

There are potentially interesting applications for these technologies, but I suspect the cost to implement them into CGC's workflows would be high.

A lot of AI/ML solutions look impressive, but they only get you about 80% accuracy at best. Unless you are implementing the solution at very large scale, it is difficult for a business to justify the investment.  Even if you had an AI that graded comics correctly 80% of the time, you'd still need humans overseeing everything. If you have 1000 full-time graders, you'd still need to retain at least 200 of them, plus management, training, and other overhead. Compare the cost to what you would pay to develop and maintain the technology.  It makes the most business sense if you can get rid of all your human graders, or if your business is big, it is no longer feasible to hire more graders.

Maybe this will change as the tech improves, but I suspect that's still a decade away. 

 

Agreed, the technology is still growing and much of its success is dictated by not just the strict financial investment one needs to acquire the equipment but the additional cost of training the AI and running a competent team behind it as well.

The points you mention are exactly what got me to thinking about it. Consistency and scalability are things CGC has to be thinking about quite a lot as of late. Seems like the automation offers a platform for both? Scaling a pool of graders is hard enough when you have location constraints I imagine, let alone keeping consistency amongst the pool as it grows. Small growth is manageable but with the stratospheric growth of the "graded" industry (all things not just comics) I suspect things will change to satisfy investments and what not. Managing customer expectations around quality and TATs is a tricky thing. Balancing retention, attracting new customers and profit margins is even trickier. 

All that said, I just got to thinking about it after seeing a bunch of threads lately with issues that seem like it would be hard to happen after multiple reviews by humans at different levels of the grading procedure. Was curious to know if there had been cases where some forms of automation had already been implemented 

 

[adampasz]

Process automation generally makes business sense when you need to scale exponentially. In CGC's case, I assume they're just trying to scale linearly, say from 100,000 customers to 120,000 in a year.

It would be cool if you could scan a cover, and then compare to a library of reference images to identify a book, and note defects like spine marks. I assume someone has experimented with this, but it's tough for me to imagine getting it to a reliable level with current technologies.

[Tony S]

PSA made headlines when they purchased Genamint to acquire their card grading AI software last year. PSA says they intend to use Genamint AI to assist with card grading.  Humans will still grade the card, but the AI software will check for trimming and other restoration. 

A couple of small card grading start ups using exclusively using AI to grade have entered the market - but are not being well received. AGS and Card-Boss are scan and upload services that then give you a grade.  A few articles are out there where graded cards were scanned and the grades were not close.  And of course you are not sending your cards in so there is no encapsulation.  

[bronze_rules]

This has been discussed on the boards before. One big issue as I see it is lack of examples to learn from. That's where CGC database could shine.

[StillOnly25Cents]

discussed a bit starting around the 43:00 mark

https://youtu.be/06xL-81x9Ro

[Doctor Dositheus]

Last time I brought up AI grading the thread got poofed.

[707comics]

Thanks for that. Interesting discussion .... the desire is certainly there. Reading between the lines it seems like the grading process might be the last leg of the process that it would apply to but other things are "lower hanging" fruit. 

[707comics]

On 10/2/2022 at 11:49 AM, Doctor Dositheus said:

Last time I brought up AI grading the thread got poofed.

Interesting. Matt is openly talking about the interesting nature of AI and how it might apply to the CGC workflow in the video shared above FWIW. 

[Sigur Ros]

On 9/29/2022 at 3:30 PM, 707comics said:

It seems with the "checks & balances" that are in place during the grading process there would be a very low chance that 2 or 3 humans looking at the book during the acceptance, grading, QA and shipping process 

I think you're misinformed.

CGC doesn't state there are "2 or 3 humans looking at the book during the acceptance, grading, QA and shipping process".

They state "The collectible has been inspected by at least two professionals"

It is a very vague statement, but I have to assume inspection applies to grading.