What Is A Robbery? Devin Haney - Vasyl Lomachenko Scored By Unbaised Artificial Intelligence.

[Cypocryphy]

ive just sent this copy of the fight to haneys pops, expect a total meltdown very shortly

Your replies and posts tell a different story mate.

[Smash]

Your replies and posts tell a different story mate.

do they?

[Cypocryphy]

ive just sent this copy of the fight to haneys pops, expect a total meltdown very shortly

lol You did? Honestly, I imagine he'll just deny it. He said that the fight wasn't even close afterward.

Did you see the Canelo and Golovkin II fight? That one was bad man. Really bad. I think if Canelo saw that, he'd crumple to the ground on the spot.

But how did you send it to Bill Haney?

[Smash]

But how did you send it to Bill Haney? [/FONT][/SIZE]

i used the internet to do that

[Cypocryphy]

do they?

Yeah. Because this is to show how DeepStrike scored controversial fights. I only have so many examples because Jabbr hasn't put out that many. So this is one of three examples, but it's getting these guys to completely overlook the AI and just focus on the Devin Loma fight and that I posted it. No mention of DeepStrike, just telling me to "move on" from this fight.

It's obvious that these guys are triggered by this fight because of repeatedly overlooking the subject of the thread, which is using AI to score fights.

[Cypocryphy]

Is the program using a 3D video? Otherwise I'm wondering how it could determine movement back and forth. Also, how can it measure "High Power commit value"? At least a human observer on the outside of the ring could see the movement and hear the impact of punches.

You'll have to ask the developers. They have a YouTube channel and are active (occasionally) on other social media, such as Reddit and X. I'd be curious as well. But I did try to really find mistakes in how it determines how hard a punch is when it lands. Apparently these guys are working with thousands of people in the industry to make this as accurate as possible and it shows. It was incredibly accurate when determining how hard a punch is when it lands.

This fight is not the best to showcase that, but the Canelo and Golovkin fight is. I recommend watching that one and assessing whether it's accurately determine the level of impact on a punch. I personally think it's remarkable. I've been mesmerized by this AI's capabilities for the last couple days.

[crimsonfalcon07]

Yeah. Because this is to show how DeepStrike scored controversial fights. I only have so many examples because Jabbr hasn't put out that many. So this is one of three examples, but it's getting these guys to completely overlook the AI and just focus on the Devin Loma fight and that I posted it. No mention of DeepStrike, just telling me to "move on" from this fight.

It's obvious that these guys are triggered by this fight because of repeatedly overlooking the subject of the thread, which is using AI to score fights.

You know I think the Haney-Loma fight was a robbery too, BUT the thread title isn't "using AI to score fights". It's functionally "Haney v Loma was a robbery because AI says so." So the criticism you're getting here is valid. TBH, it makes the other fights you posted look like a show just to try to make your narrative here look better. Bear in mind this is coming from someone who largely agrees with the premise too.

The point of this thread is to prove that Haney v Loma was a robbery. And I don't buy that Loma won 10 rounds either. I've watched that fight several times in slow-mo as well, so I don't buy the premise that the AI is as good as it claims. I know for a fact that Loma tends to vary the power behind his punches to land the ones that matter. So this seems to have a tendency to equate a pitter pat distraction punch with a real punch because it can't tell the difference. I'll count it as a scoring punch, but there's more to it.

As far as the notion of scoring fights by AI goes, I think it's flawed from the outset, and moreover that it won't be adopted unless it's already been corrupted. There's a lot of things that on paper says AI should be better than a human, but the practice hasn't turned out that way in any field yet.

[Cypocryphy]

You know I think the Haney-Loma fight was a robbery too, BUT the thread title isn't "using AI to score fights". It's functionally "Haney v Loma was a robbery because AI says so." So the criticism you're getting here is valid. TBH, it makes the other fights you posted look like a show just to try to make your narrative here look better. Bear in mind this is coming from someone who largely agrees with the premise too.

The point of this thread is to prove that Haney v Loma was a robbery. And I don't buy that Loma won 10 rounds either. I've watched that fight several times in slow-mo as well, so I don't buy the premise that the AI is as good as it claims. I know for a fact that Loma tends to vary the power behind his punches to land the ones that matter. So this seems to have a tendency to equate a pitter pat distraction punch with a real punch because it can't tell the difference. I'll count it as a scoring punch, but there's more to it.

As far as the notion of scoring fights by AI goes, I think it's flawed from the outset, and moreover that it won't be adopted unless it's already been corrupted. There's a lot of things that on paper says AI should be better than a human, but the practice hasn't turned out that way in any field yet.

The only thing true about what you said is that you believe I posted this to push some agenda on the Loma Haney fight, that that is your perception. That's really what you're saying here because the rest is not true.

This is the problem when people take things out of context. I published the Canelo and Golovkin fight, and in that thread, and forum member asked that if there were any other controversial fights. So when I went to look, I found the Loma fight, which is the only other controversial fight on there (unless you consider Hagler and Duran controversial). And I posted it on that thread.

I posted it in part because yesterday (the day I posted this thread), Brad Goodman was on the ThaBoxingVoice, and he said he rewatched and rescored the fight. He said at first he thought it was a robbery but now he thinks that Moretti's scorecard was accurate. (That's right, Moretti's scorecard of 116-112) So in light of these recent comments by Brad Goodman, I made this post asking "What is a robbery" or "Was this a robbery." And to watch "AI score the fight." I didn't expect to trigger so many people considering so much time has passed, although I suppose I should have expected it.

As to what else you said, I think you need to step away from your own scorecard and try to see this from an unbiased, unemotional perspective. I challenge you to find something the AI missed or did not assess correct. And in regard to Loma varying the strength of his punches, if it lands on Haney, even just a touch jab, it is still a scoring blow. It would not be a significant punch, but it is supposed to be factored by the judges. So I'm not sure what you're saying here.

I believe a lot of you don't understand how far AI has evolved and how efficient DeepStrike is at assessing who lands what and how hard it lands. So I challenge you to find a mistake by DeepStrike and post it here. Personally, I only found one mistake in that entire fight. That's it. And I tried to find as many as possible the other day.

And in regard to the last point, if this is flawed, it doesn't change the fact that it is far superior to what is being used presently. FAR SUPERIOR. Using this would be a massive step in the right direction because I would rather have a 99.99 percent error rate than a 70 to 80 percent error rate, which is approximately what we have presently.

[Zelda]

Artificial intelligence assessed the fight, and tallied all the punches. It was pretty clear that Loma won 10 rounds. By your standards, that would be considered a robbery. You can watch how it was scored.

I didn't have time to go through the entire fight but judging by the first two rounds, that AI has a lot of work to do before scoring a fight. It appears if the fighter throws a punch, it is counted even if it does not or barely touches an opponent. Boxing is scored in a very complicated and subjective way. It would be interesting to know what the AI is trained on...past fights and scorecards, human annotation on a punch-by-punch basis, com*****g punches thrown, or what?

From the way it was describing the shorts, it appears it is based on a computer vision dataset that tries to identify the action of throwing a punch rather than one that landed or caused any damage. That is why I say it has some way to go because that system can actually mimic human scoring. I am not sold out on this as of yet.

[Cypocryphy]

I didn't have time to go through the entire fight but judging by the first two rounds, that AI has a lot of work to do before scoring a fight. It appears if the fighter throws a punch, it is counted even if it does not or barely touches an opponent. Boxing is scored in a very complicated and subjective way. It would be interesting to know what the AI is trained on...past fights and scorecards, human annotation on a punch-by-punch basis, com*****g punches thrown, or what?

From the way it was describing the shorts, it appears it is based on a computer vision dataset that tries to identify the action of throwing a punch rather than one that landed or caused any damage. That is why I say it has some way to go because that system can actually mimic human scoring. I am not sold out on this as of yet.

Can you timestamp some examples from the first two rounds? That would really help support what you are saying in your response. The AI measures impact into five categories ... actually, six categories. It measures and classifies the punches into six categories:

1. missed punch
2. Minimum punch
3. Low impact punch
4. Middle impact punch
5. High impact punch
6. Maximum impact punch

It appears to me that you haven't really looked to see what it's doing because your response doesn't reflect what's actually being done. For example, here's round 1 and round 2 for both fighters.


Round 1 Haney Impact Category

Max	0
High	1
Mid	2
Low	1
Min	1
Miss	31

Round 1 Loma Impact Category

Max	0
High	0
Mid	0
Low	2
Min	2
Miss	26

​

Round 2 Haney Impact Category

Max	0
High	1
Mid	3
Low	1
Min	6
Miss	32

​

Round 2 Loma Impact Category

Max	0
High	3
Mid	2
Low	1
Min	8
Miss	47

​

So you can see that it is separating out punches on a six factor scale. But like I said, if you can timestamp where you see a mistake or issue, that would definitely help.