Will Scotland Yards HOLMES 2 and AI solve Jack the Ripper?

[Lewis C]

But who knows if he was or not? Kosminski, Kaminski, Cohen, Levy, Hyams, Chapman...all Jewish suspects.

I believe that Chapman was a Gentile.

[Fiver]

I would suggest Lawende is giving an honest statement of what he remembers seeing and that all three saw a couple, who was probably Kate and her killer given the location and timings. Could this description stand up in court, of course not.

Paul

It is probable that Lawende gave a description that he believed to be correct. But we don't have his direct statement, we have newspaper accounts and police reports and they contradict each other on the suspect's height, hair color, mustache size, and how well they were dressed. We can try to sort these contradictions out, but we lack a signed statement from Lawende and cannot interview him to obtain clarification.

Both at the time and today, people tend to favor Lawende over Levy and Harris, because of the level of detail. (Ironically, many discard Hutchinson because of his level of detail, even though Hutchinson had a lot more time to observe his suspect.) Lawende's description can be used for and against suspects or at least types of suspects. Levy and Harris are of little or no use for identifying a suspect.

There are two possibilities - either Lawende had significantly better night vision and memory than his companions or he had an overactive imagination. Harris clearly supports the latter as if the man's back was turned, Lawende could not have seen a kerchief or mustache. Levy clearly believed Lawende was very wrong about the suspect's height, leaving open the possibility Lawende was wrong about other elements of his description.

There's also the issue of Lawende identifying Eddowes. She was a stranger seen for only a few moments in very poor lighting. If Lawende was right, the man he was was the Ripper. If Lawende was wrong, and the woman was not Eddowes, then the man was not the Ripper and the accuracy of his description is irrelevant.

Church Lane was not the only way to get to Mitre Square. If Lawende misidentified the woman, then Eddowes and her killer entered through one of the other two routes and may have already been in Mitre Square at 1:35am.

[Fiver]

Lechmere would. Lives outside the boundary, but for real-world reasons connected with home, family and work is actually confined within it at certain times, and with very limited degrees of freedom.

Applying a silly binary model to this real-world life generates the wrong answer of a Marauder based near Tabram's murder site. Why even bother with it?

M.

There are reasons to question the Commuter-Marauder model. I have not seen enough information to tell if its a useful model or a waste of time like the organized-disorganized dichotomy. We also have no information as to how they came to the Marauder conclusion.

But you aren't discarding the answer based on criticism of the model or the reasoning used to reach that answer. You are discarding the answer because it rules out your suspect.

[John Wheat]

There are reasons to question the Commuter-Marauder model. I have not seen enough information to tell if its a useful model or a waste of time like the organized-disorganized dichotomy. We also have no information as to how they came to the Marauder conclusion.

But you aren't discarding the answer based on criticism of the model or the reasoning used to reach that answer. You are discarding the answer because it rules out your suspect.

Nothing at all rules Lechmere in though. He's a witness not a suspect.

[bonestrewn]

The use of AI in this case brings up two things for me:

1. It has been demonstrated that AI systems repeat and magnify the prejudices embedded in the data on which they are trained. For example, giving an image generating AI the prompt "doctor" and receiving images of only men back. Or facial recognition software which, due to embedded and unrecognized racial biases, is trained primarily on images of white people, then struggles to recognize people of color.

When we talk about the evidence and reporting in this case, we are talking about information deeply affected by xenophobia and antisemitism. This isn't to say the information has zero value, but that an AI given this information will not have the human discernment to determine relative value. I am not convinced that if we gave an AI all the Ripper documentation, it wouldn't spit out a caricature inflected by these biases.

2. This is more of a personal theory of mine, but I do not feel confident that any serious model of "serial killing" developed post-Ripper has almost anything of value to give. I feel that the Ripper has completely poisoned the well by "teaching" other serial killers how to act, and so to draw any conclusions about serial killers and their behavior post-1888, you are necessarily working with a tainted pool of information that, again, requires human discernment to sieve for meaningful info. So, to argue that this system or integrated platform has been taught models and structures based on the modern study of serial killers does not convince me that it will have anything useful to say.

[Patrick Differ]

The use of AI in this case brings up two things for me:

1. It has been demonstrated that AI systems repeat and magnify the prejudices embedded in the data on which they are trained. For example, giving an image generating AI the prompt "doctor" and receiving images of only men back. Or facial recognition software which, due to embedded and unrecognized racial biases, is trained primarily on images of white people, then struggles to recognize people of color.

When we talk about the evidence and reporting in this case, we are talking about information deeply affected by xenophobia and antisemitism. This isn't to say the information has zero value, but that an AI given this information will not have the human discernment to determine relative value. I am not convinced that if we gave an AI all the Ripper documentation, it wouldn't spit out a caricature inflected by these biases.

2. This is more of a personal theory of mine, but I do not feel confident that any serious model of "serial killing" developed post-Ripper has almost anything of value to give. I feel that the Ripper has completely poisoned the well by "teaching" other serial killers how to act, and so to draw any conclusions about serial killers and their behavior post-1888, you are necessarily working with a tainted pool of information that, again, requires human discernment to sieve for meaningful info. So, to argue that this system or integrated platform has been taught models and structures based on the modern study of serial killers does not convince me that it will have anything useful to say.

Artificial Intelligence is only one component of Big Data Analytics. Query based analysis saves time and creates other avenues based on all types of data including perhaps the most important in this case, the medical data , MO and signatures.

The University of Washington did a Ripper study and compared it to over 3000 other murders, including by serial killers. One conclusion they had was that the mutilations by this killer were extremely rare. Even to this day it just rarely, if
ever, happens. They also concluded that the killer was a local.

I haven't seen any profiles of any kind or studies indicating this killer was a CCommuter.if there are I would highly appreciate being pointed in that direction.

If you have more tools to investigate or create new avenues in pursuit of answers or create new questions then I see no reason to not use it. It would be nice to see a Big Data analysis of the medical evidence that is the only constant in the case.

Of course none of us have seen

[Patrick Differ]

Continued ( fat fingers)
None of us have seen the data sets but as someone pointed out is there enough information in the case to gain anything useful?

The Who in this case is the big question. The second in my mind might be who had the ability to commit the rare act of mutilating a human. Would that eliminate those who had no known skills with a knife, for example?

So if Big Data Analytics produces a result based on serial killing or murder information residing in a database, literally thousands of records, and statistically says these mutilations are rare...then how do you use that information?

Ask more questions?
Why was it rare?
Did it require a certain skill to make the cuts?
Did the knife, based on the evidence, have to be a certain sharpness to penetrate human skin?

For me I am interested in the questions that have not been asked or fully explored. If any at this point. That's why I would not discount big data based on large historical samples.

One of these might be- How would someone mutilate ( based on the medical evidence) really perform this act in less than 10 minutes, in 5 minutes? How sharp did the knife have to be each time relative to the previous mutilation? How did human skin differ from animal skin? And so on?

Can Big Data get us closer? I don't think it can hurt is my only point. The JtR mutilations were rare statically. Does that tell us something about the killer? Something in his profile?

[Richard Patterson]

What is HOLMES 2 ? Basically it is a Platform for Big Data Analytics. It is based on Palantir Technologies and likely other apps integrated for, in this case, criminal analysis. Scotland Yard, FBI..most Agencies have similar Platforms today.

I didn't know anything about the movie..Jack the Ripper..Case Reopened with Emilia Fox and David Wilson, who in my mind, were more Moderators than Experts. I just watched it. I did search Emilia Fox and saw, and read, the Pub Talk on Wilson. I cant comment on Wison or Fox, I really don't know of them but I did research Scotland Yards HOLMES and HOLMES2. I recognized Palantir immediately from my previous life before retiree. I could find nothing in the search on HOLMES.

What struck me is that Scotland Yard had an overall " Objective " analysis using HOLMES and this was very different from the " subjective" analysis and rankings ive seen attempted on Casebook. My apologies if an Objective Analysis exists on the site outside of what Scotland Yard did.

To clarify, the input data was Objective data based on sworn testimony, records, photographs and any evidence Scotland Yard determined to be authentic. So an example might be that the sworn testimony of eyewitnesses would be considered factual. Unfortunately its not clear how far Scotland Yard has gone with the analysis. They did share some things.

There were some findings by Experts in fields like Geospatial modeling , forensics and psychological profiling that were applied. Here are 5 that I found interesting:

1. Martha Tabrum was the first victim. Determined by overkill and method.
2. JtR was a Marauder not a Commuter- meaning he lived in the area.
3. The next 4 victims Nichols, Chapman, Stride and Eddowes were virtually of equal distance from Tabrum in the East, North, South and West Geospace.
4. Mary Kelly was the only victim murdered inside and only victim to be skinned.
5. All 6 victims were murdered in a confined space.

So the Rippers boundaries in each direction are his comfort zone. Scotland Yard drew horizontal lines through Chapman and Stride and Vertical lines through Eddowes and Nichols. This was the overall zone of where JtR felt comfortable traveling.

The other finding in the HOLMES analysis was a heat map that showed higher probabilities of comfort zone. The highest marked by proximity of Tabrum to Chapman, Eddowes and Kelly. This was also close to Butchers Row and basically in the Jewish Quarter.

The Scotland Yard conclusion, so far, is that the killer lived near where Martha Tabrum was murdered and this was the killers first victim. They select Aaron Kosminski as the likely killer because of his profile and where he lived. However it's not clear if Kosminski lived on Greenfield St or Providence Ct.

In truth, the other candidate that fits the Scotland Yard analysis using Objective data, is of course Jacob Levy. Both he and Kosminski lived near George Yard. Jacob Levys wife actually lived right there on Bull CT. Levy also lived on Fieldgate Street near Greenfield.

I can't imagine that Scotland Yard would not use every bit of factual data it could get it's hands on. I know they have the best modern tools for analysis and would consider that an advantage.

If JtR wasn't Kosminski or Levy then it was likely someone unknown, perhaps a yet to be identified slaughterman on Butchers Row that lived with his brother. Kosminski lived with his brother and Levy was a butcher. Both insane.

HOLMES2? what next? A.I. ?

Richard Patterson — Researcher

What you describe with HOLMES 2 is exactly what fascinates me about AI in historical cases: it forces you to separate documented descriptors from interpretation. Scotland Yard can feed sworn testimony, police reports, maps, asylum records, etc., into a system, and the machine will simply cross-check. No drama, no bias.

That’s what I’ve been experimenting with on the Ripper suspects. I’ve written a strict search prompt for AI that does the same thing — it only allows five descriptors taken from Major Henry Smith’s memoir (ex-medical student, asylum history, associations with prostitutes, petty fraud, Haymarket presence). Then I lock in the suspects named in the Macnaghten memorandum and compare them. The AI isn’t told who to pick. It just runs the descriptors against the historical record and counts matches.

What’s striking is that when you make the machine work only with documented facts (no speculation, no “feel” for a suspect), it keeps putting Francis Thompson at the front of the line. Not because I asked it to, but because that’s how the evidence stacks. Once an AI has run that kind of table, it won’t “forget” the result just because someone argues loudly for another suspect — the numbers keep pulling it back.

To me, that’s the real strength of AI here. It’s not about replacing researchers. It’s about building an objective comparison layer that everyone can test and audit. Like HOLMES 2, it doesn’t care about reputation or tradition; it just adds up what’s documented.

Of course, human judgment still matters — we decide which descriptors to test, and whether the sources are solid. But once the framework is there, the computer won’t fudge or play favourites. That’s what makes me think this kind of structured AI search will become an essential tool in historical crime analysis, right alongside the GIS mapping and forensic review you mentioned.

[Herlock Sholmes]

Richard Patterson — Researcher

What you describe with HOLMES 2 is exactly what fascinates me about AI in historical cases: it forces you to separate documented descriptors from interpretation. Scotland Yard can feed sworn testimony, police reports, maps, asylum records, etc., into a system, and the machine will simply cross-check. No drama, no bias.

That’s what I’ve been experimenting with on the Ripper suspects. I’ve written a strict search prompt for AI that does the same thing — it only allows five descriptors taken from Major Henry Smith’s memoir (ex-medical student, asylum history, associations with prostitutes, petty fraud, Haymarket presence). Then I lock in the suspects named in the Macnaghten memorandum and compare them. The AI isn’t told who to pick. It just runs the descriptors against the historical record and counts matches.

What’s striking is that when you make the machine work only with documented facts (no speculation, no “feel” for a suspect), it keeps putting Francis Thompson at the front of the line. Not because I asked it to, but because that’s how the evidence stacks. Once an AI has run that kind of table, it won’t “forget” the result just because someone argues loudly for another suspect — the numbers keep pulling it back.

To me, that’s the real strength of AI here. It’s not about replacing researchers. It’s about building an objective comparison layer that everyone can test and audit. Like HOLMES 2, it doesn’t care about reputation or tradition; it just adds up what’s documented.

Of course, human judgment still matters — we decide which descriptors to test, and whether the sources are solid. But once the framework is there, the computer won’t fudge or play favourites. That’s what makes me think this kind of structured AI search will become an essential tool in historical crime analysis, right alongside the GIS mapping and forensic review you mentioned.

So that’s how you arrived at the erroneous match. Someone on here said that they sensed AI in your work. Like I said before…you plug in incorrect date and what do expect? You just stated that you added ‘petty fraud,’ for Thompson. How is finding 2 coins ‘petty fraud?’ It’s not. Not by anyone’s thinking except yours. It’s proven beyond any doubt at all that Smith’s suspect was Oswald Puckridge

Medical student - exact match.
Lunatic asylum - exact match.
Rupert Street - exact match.

What are the chances of there being another ex-medical student, recently released from an asylum living in the exact street that Smith said that his suspect would be found in (and low and behold…there he was…with polished farthings on him) No one but you in the world would favour Thompson over Puckridge on this. Your position is bizarre to put it mildly.