noelward
Well-known member
Why AI is a Bit Like Having Kids (part 1)
What matters is the training
By Noel Ward, Editor@Large
Artificial Intelligence is getting a lot of headlines but that doesn’t mean much. Blah, blah, blah. Yadda, yadda, yadda.
What matters is how you can use it as a printer. You are probably already using some of AI’s most basic forms…
Is it intelligence?
Human intelligence, claims Britannica, combines learning, reasoning, problem-solving, perception, and use of language. It does not include educated guesses or mention that some animals—especially wild ones—are as intelligent as they need to be. Artificial intelligence works without any gray matter, substituting a mix of rules and instructions, called algorithms that tell inanimate devices to do things that make them appear smart. Glowing or ominous headlines aside, consider how your phone’s autocorrect and voice recognition apps offer up inscrutable errors. They are a hint that AI has a way to go.
The Enablers
Those algorithms are enablers. Wikipedia calls algorithms a sequence of instructions for solving specific problems or performing computations. The magic arrives when algorithms team up to create what appears to be ‘intelligence.’ Among the more powerful algorithms are those of ‘Generative AI,’ a customizable variant usually abbreviated as ‘Gen AI.’ This often draws on the LLM or Large Language Model, often partnering with Natural Language Processing or NLP. These are part of how AI software is “trained” on topics and how words go together. The resulting statements make AI apparently understand questions. It does not actually do that, but the illusion is compelling. Testing this out, I had a conversation with one flavor of AI. I thought it akin to a superficial cocktail party conversation with a stranger but what was eerie is that its responses improved as I fed it more info. Hmmm. The geeks in the Silicon Valley may be onto something.
Gen-AI is further enhanced by knowledge retrieval and response generation making it a key driver of AI growth. In a 2023 Forbes research study, a third of respondents’ organizations said they were using Gen AI in one or more business functions. More telling is the four in ten who said their companies’ AI investments are likely to increase because of Gen AI. Being skeptical by experience, this makes me worry.
So what does this have to do with printing?
Good question. I told you we’d get here. Thanks for hanging in.
Are you using AI in the ink-on-paper part of your business? It’s okay if your answer is, “it depends.” As I discovered when installing a new office printer, default settings use AI to inform of print volumes, supply levels, machine performance, and maintenance needed. This is the new normal for digital presses or digitally enabled offset presses, much in the way your car may alert you of pending service.
For example, AI may alert a press operator to upcoming scheduled maintenance intervals or warn that the cyan inkjet printhead is randomly misfiring, that the device cannot correct for the error, and a part you don’t have is required. It then may alert a vendor service tech and inform the machine operator of the pending downtime. This can be especially useful if the service will put the press offline for three hours. Meanwhile, the sole operator of your two sheet-fed offset presses might learn of a problem that needs attention on the sixth color station of one press and that paper is running low on the second one. One person for two offset presses? Yep. Coming soon to a shop floor near you.
“Reducing tasks lets people and devices perform more business-sensitive activities with a focus on revenue generation,” explains Jenna Miner, Channel Development Manager at ConnectWise, a software company focusing on IT solutions.
Some print providers already use AI for customer follow-up and prospecting. Others use it to improve the customer experience with chatbots that gather data for analysis and aid machine learning. Some equipment vendors use AI to help find qualified labor. Still other AI versions are giving technicians immediate access to machine operability and recommended fixes. If you self-maintain, this puts OEM-level knowledge in the hands of your press operators. Note that this is (probably) an accurate version of repair, not the usual You Tube junk. AI can’t do all this right now, but it will before long.
Curious, I pulled back the curtain
There were pages of algorithms. High-end math and indecipherable Python and C++ code. Not my wheelhouse.
So I asked OpenAI’s ChatGPT 3.5 and Google’s Gemini (recently name-changed from Bard) the same broad question: “How can commercial printers use AI?”
Both AI engines scrape the web for data, and work in similar ways. I thought the similar answers provided were more suggestions than useful. Further probing probably helps, as would being brand-and-model specific. Responses reflected basic answers of software and equipment vendors, industry analysts, and contributors to this website. While the responses lacked a context print providers could relate to there were still valid points. For example, AI suggested print providers use market segmentation, various advertising and marketing options, customer relations, and staffing to help improve a business. Not to be outdone, equipment and software vendors probably use AI when defining tactics they think should be applied to your operation.
So I went more brand specific. “How can a commercial printer with a [name of] press be more competitive? I got back more generic responses that could apply to any press but many of the responses could still be acted on or more deeply queried without a lot of imagination.
Hoping for more I asked, “How can I solve printhead problems with my [name of] press?” I was given a standardized list of troubleshooting steps and was encouraged to call the vendor for support. This was followed by the message, “Unfortunately, I cannot offer specific troubleshooting advice without knowing the exact model of your [name of] press and the specific printhead problems you're experiencing.” Hmmm, there may be hope. If endors are not doing so already, thry will soon offer up branded, AI-enabled customer portals. Customer support sites may become AI chat portals with behind-the-scenes reporting. That may or may not improve the customer experience.
Next, I asked about how salespeople should be selling the UV curing capability of an offset press. Although my question was deliberately broad, this response was better, citing the importance of educating both customers and salespeople on the advantages of UV printing and curing. Not bad. And not inaccurate.
On the upside, all the answers came back in seconds. I quickly realized I could do this all day—as I’m sure you could. It was fun and sure beat Googling and clicking links.
My initial foray was with ChatGPT 3.0. ChatGPT 3.5 was better. Both are free. ChatGPT 4.0 is much better and still cheap at $20 a month (although it arrives with disclaimers and legal agreements). I suspect the 4.0 version may be free, or close to it, later this year.
Programmed to admit
The generic summaries I received from both ChatGPT and Gemini were a useful “State of AI” measurement, albeit replete with disclaimers. Both claimed to be only ‘partially trained.’ OpenAI warned, “ChatGPT can make mistakes” and to “Consider checking important information.” Gemini advised that it, “…may display inaccurate info, including about people, so double-check responses.” In other words, taking what AI says straight to your financial guys might not be the best move. Both said AI is “programmed to admit that acting on its advice requires human judgement.” (Note the “programmed to admit” part). On the other hand, all sources should be checked whether the answer is arrives on your laptop or comes from a human sitting across the desk.
However.
The AI landscape is changing fast. Tech Republic reported on February 23 that Google’s release of Gemma, a lighterweight version of Gemini, uses a form of Gen-AI. It is said to be able to run on a workstation and be flexible enough to let organizations build custom bots without the workload and expertise needed for fine-tuning more capable applications. The workstation part means AI you can fiddle with may be coming soon to a computer near you. Maybe like Excel or QuickBooks. What can possibly go wrong?
For now, extracting AI’s power requires knowing where and how it can add value for your business and customers. Be skeptical and know its limitations. Ask it smart questions and probe for more. Just as email, websites and web searches have become business as usual, AI will become normal. It will probably be pretty solid once it gets smarter. For now, AI a bit like having kids. You need to nurture, train and push a little while it’s young... so you can shake your head later.
What matters is the training
By Noel Ward, Editor@Large
Artificial Intelligence is getting a lot of headlines but that doesn’t mean much. Blah, blah, blah. Yadda, yadda, yadda.
What matters is how you can use it as a printer. You are probably already using some of AI’s most basic forms…
- Your digital printers tell you when they are low on toner or ink
- Movies and TV shows are urged upon you by streaming services.
- Amazon suggests items you don’t want based on previous purchases.
- Your robo-vacuum cleaner is freaking out your dog… while it maps your house.
- A chatbot solved a simple problem while a human was “busy servicing other customers.”
Is it intelligence?
Human intelligence, claims Britannica, combines learning, reasoning, problem-solving, perception, and use of language. It does not include educated guesses or mention that some animals—especially wild ones—are as intelligent as they need to be. Artificial intelligence works without any gray matter, substituting a mix of rules and instructions, called algorithms that tell inanimate devices to do things that make them appear smart. Glowing or ominous headlines aside, consider how your phone’s autocorrect and voice recognition apps offer up inscrutable errors. They are a hint that AI has a way to go.
The Enablers
Those algorithms are enablers. Wikipedia calls algorithms a sequence of instructions for solving specific problems or performing computations. The magic arrives when algorithms team up to create what appears to be ‘intelligence.’ Among the more powerful algorithms are those of ‘Generative AI,’ a customizable variant usually abbreviated as ‘Gen AI.’ This often draws on the LLM or Large Language Model, often partnering with Natural Language Processing or NLP. These are part of how AI software is “trained” on topics and how words go together. The resulting statements make AI apparently understand questions. It does not actually do that, but the illusion is compelling. Testing this out, I had a conversation with one flavor of AI. I thought it akin to a superficial cocktail party conversation with a stranger but what was eerie is that its responses improved as I fed it more info. Hmmm. The geeks in the Silicon Valley may be onto something.
Gen-AI is further enhanced by knowledge retrieval and response generation making it a key driver of AI growth. In a 2023 Forbes research study, a third of respondents’ organizations said they were using Gen AI in one or more business functions. More telling is the four in ten who said their companies’ AI investments are likely to increase because of Gen AI. Being skeptical by experience, this makes me worry.
So what does this have to do with printing?
Good question. I told you we’d get here. Thanks for hanging in.
Are you using AI in the ink-on-paper part of your business? It’s okay if your answer is, “it depends.” As I discovered when installing a new office printer, default settings use AI to inform of print volumes, supply levels, machine performance, and maintenance needed. This is the new normal for digital presses or digitally enabled offset presses, much in the way your car may alert you of pending service.
For example, AI may alert a press operator to upcoming scheduled maintenance intervals or warn that the cyan inkjet printhead is randomly misfiring, that the device cannot correct for the error, and a part you don’t have is required. It then may alert a vendor service tech and inform the machine operator of the pending downtime. This can be especially useful if the service will put the press offline for three hours. Meanwhile, the sole operator of your two sheet-fed offset presses might learn of a problem that needs attention on the sixth color station of one press and that paper is running low on the second one. One person for two offset presses? Yep. Coming soon to a shop floor near you.
“Reducing tasks lets people and devices perform more business-sensitive activities with a focus on revenue generation,” explains Jenna Miner, Channel Development Manager at ConnectWise, a software company focusing on IT solutions.
Some print providers already use AI for customer follow-up and prospecting. Others use it to improve the customer experience with chatbots that gather data for analysis and aid machine learning. Some equipment vendors use AI to help find qualified labor. Still other AI versions are giving technicians immediate access to machine operability and recommended fixes. If you self-maintain, this puts OEM-level knowledge in the hands of your press operators. Note that this is (probably) an accurate version of repair, not the usual You Tube junk. AI can’t do all this right now, but it will before long.
Curious, I pulled back the curtain
There were pages of algorithms. High-end math and indecipherable Python and C++ code. Not my wheelhouse.
So I asked OpenAI’s ChatGPT 3.5 and Google’s Gemini (recently name-changed from Bard) the same broad question: “How can commercial printers use AI?”
Both AI engines scrape the web for data, and work in similar ways. I thought the similar answers provided were more suggestions than useful. Further probing probably helps, as would being brand-and-model specific. Responses reflected basic answers of software and equipment vendors, industry analysts, and contributors to this website. While the responses lacked a context print providers could relate to there were still valid points. For example, AI suggested print providers use market segmentation, various advertising and marketing options, customer relations, and staffing to help improve a business. Not to be outdone, equipment and software vendors probably use AI when defining tactics they think should be applied to your operation.
So I went more brand specific. “How can a commercial printer with a [name of] press be more competitive? I got back more generic responses that could apply to any press but many of the responses could still be acted on or more deeply queried without a lot of imagination.
Hoping for more I asked, “How can I solve printhead problems with my [name of] press?” I was given a standardized list of troubleshooting steps and was encouraged to call the vendor for support. This was followed by the message, “Unfortunately, I cannot offer specific troubleshooting advice without knowing the exact model of your [name of] press and the specific printhead problems you're experiencing.” Hmmm, there may be hope. If endors are not doing so already, thry will soon offer up branded, AI-enabled customer portals. Customer support sites may become AI chat portals with behind-the-scenes reporting. That may or may not improve the customer experience.
Next, I asked about how salespeople should be selling the UV curing capability of an offset press. Although my question was deliberately broad, this response was better, citing the importance of educating both customers and salespeople on the advantages of UV printing and curing. Not bad. And not inaccurate.
On the upside, all the answers came back in seconds. I quickly realized I could do this all day—as I’m sure you could. It was fun and sure beat Googling and clicking links.
My initial foray was with ChatGPT 3.0. ChatGPT 3.5 was better. Both are free. ChatGPT 4.0 is much better and still cheap at $20 a month (although it arrives with disclaimers and legal agreements). I suspect the 4.0 version may be free, or close to it, later this year.
Programmed to admit
The generic summaries I received from both ChatGPT and Gemini were a useful “State of AI” measurement, albeit replete with disclaimers. Both claimed to be only ‘partially trained.’ OpenAI warned, “ChatGPT can make mistakes” and to “Consider checking important information.” Gemini advised that it, “…may display inaccurate info, including about people, so double-check responses.” In other words, taking what AI says straight to your financial guys might not be the best move. Both said AI is “programmed to admit that acting on its advice requires human judgement.” (Note the “programmed to admit” part). On the other hand, all sources should be checked whether the answer is arrives on your laptop or comes from a human sitting across the desk.
However.
The AI landscape is changing fast. Tech Republic reported on February 23 that Google’s release of Gemma, a lighterweight version of Gemini, uses a form of Gen-AI. It is said to be able to run on a workstation and be flexible enough to let organizations build custom bots without the workload and expertise needed for fine-tuning more capable applications. The workstation part means AI you can fiddle with may be coming soon to a computer near you. Maybe like Excel or QuickBooks. What can possibly go wrong?
For now, extracting AI’s power requires knowing where and how it can add value for your business and customers. Be skeptical and know its limitations. Ask it smart questions and probe for more. Just as email, websites and web searches have become business as usual, AI will become normal. It will probably be pretty solid once it gets smarter. For now, AI a bit like having kids. You need to nurture, train and push a little while it’s young... so you can shake your head later.