Artificial intelligence has become a mainstream topic of conversation, thanks mainly to concerns about text generation software being able to create student essays indistinguishable from those written by two-thumbed students.
This is the Large Language Model flavour of AI where hugely powerful computers and algorithms have been able to parse everything written and published on the internet, have learned the probability of one word following another according to the context in which it is being used, and can construct sentences that follow a logical structure.
The result is that the software can deliver an academic essay, a proposal, an email, thank you note and as part of the campaign to demonstrate the power of the applications, what passes as a poem. And all in seconds and with almost no effort.
ChatGPT has garnered much of the attention, but other LLMs, as they say, are available and many more will be, some better and many worse than those released by Google or Microsoft.
They may not, as it turns out, be winning academic qualifications for students, but they will be creating reports, writing instruction manuals and producing short articles for newspapers, mundane formulaic writing that is a chore to produce. Do not expect War & Peace, though there will certainly be books written entirely by AI. They will not be literary masterpieces or original in any way – and in all probability will never be printed for sale.
There is a key limitation to this type of AI: it draws upon what has already been published. The software is incapable of generating anything that is truly new, of making that unexpected connection that is a hallmark of true creativity. In many ways these algorithms are the next generation of search tools rather than anything more.
This is also just one, highly visible, application of artificial intelligence and one of limited direct use to printers. AI is not just about rearranging words, some are image generating programs that have been gaining attention for mixing bizarre instructions: my grandson playing in the sand in the style of Van Gogh for example – and coming up with equally bizarre images, some of which include hands with six or more fingers. One algorithm was asked to create a picture of salmon swimming up stream and did so, but using images of filleted salmon from the fishmonger’s slab rather than actual fish.
Voice AI is creating some anxiety for it will be possible to use samples of someone’s voice and simulate a conversation that might be used for fraudulent purposes. If it exists online, any content is fair game for manipulation by this technology.
Other AI algorithms are already in place. Certainly AI is already driving chatbots for low level customer support purposes. It is sure to be used in training operators to run pieces of equipment, responding to questions as they arise.
AI is also the next step forward for machine operation. Many press producers are working on applications of AI to automate makeready, sequencing process steps in the right order with no operator intervention, for maintenance where sensors are recording the status of all manner of settings throughout the press and using pattern recognition to understand when something needs attention or is coming close to the end of its life and anticipating component failure and unexpected breakdowns.
Like all AI, this demands vast amounts of data. Indeed the one defining characteristic of any AI application is the hinterland of data that comes before it. AI cannot exist without billions of data points. And any AI that tries to work with a limited data set will only be of limited or broad use. Heidelberg with thousands of presses, each printing dozens of jobs a day, each with hundreds of sensors sending readings back to a server somewhere in the cloud, in real time throughout the day, can establish the patterns that identify upcoming maintenance issues, trouble with a particular ink, the optimum speed and weights for any widely used substrate.
A company with only a handful of users and where machinery is infrequently used will not be able to do this. The depth of data is simply not there. For these companies to share in the spoils of AI, data has to be gathered and in such a way as to impose no additional actions on the company using the equipment that is being harvested for this data. On board sensors are the way to achieve this.
At the moment most of the benefit of AI accrues to the developers and suppliers. They get insight into how machines are being used in the field; they get to identify any weakness that a next iteration might iron out, and they get to manage maintenance call outs by being able to predict and schedule when maintenance is necessary.
Another instance of AI is in automatically generating simple design jobs. Many online design tools are able to create business cards, flyers, menus and more calling upon a vast database of previous designs and can filter these by plumbers, corner cafes, window cleaners and so on, to help shape designs for inexperienced or occasional print buyers. It is not design as a designer would recognise it.
Other applications are able to scale a known label or carton design to create a version specific to a new customer, who is also unlikely to be an experienced designer. Such tools are a way to cope with the deluge of demand for short run packaging that is expected as digital printing becomes established in this sector. For the moment real experience and guidance will win out over the chatbot and parametric design as it is known. These customers need confidence that human communication brings. But this may change in future.
The real gain from AI comes when these designs, whether generated by algorithms or not, reach the production process. Today’s crop of MIS providers can make a reasonable stab at scheduling, but it is limited: all jobs on this weight and type of paper; all jobs to be delivered tomorrow; all jobs needing to be folded into 16pp sections, for example.
They are not always so good at reacting to late changes, a job which is running late, a missing file or lack of paper perhaps. Because this data is lacking. The system can be interrogated after the event, when compiling the invoice say, but live data from every point in the business is vanishingly rare.
Where, for example, is information about a pallet of work in progress? It may be known that the sheets have been printed, with a barcode to help download the data to set up the folder with no further intervention, but until that barcode is scanned, the work in progress has dropped off the radar.
Press suppliers, whether analogue or digital, have concentrated on their piece of equipment not on linking up disparate devices from numerous suppliers.
Connectivity though is slowly coming, whether through the evolution of JDF, APIs or platforms like Zaikio’s Mission Control. Some say the biggest problem with this is its close relationship with Heidelberg and that company’s history of trying to keep its customers within Heidelberg’s walled garden. Hopefully that attitude no longer prevails. Even the mighty Heidelberg must recognise that it cannot dominate every aspect of the printing industry as it might have done in the past. It needs to be open, transparent and to be just as connected as its customers are to their customers.
Despite the changed circumstances, Heidelberg is still in a pivotal position and it believes that its Push to Stop and other software are key differentiating factors that drive machinery sales and keep its factory turning. This is true. But Heidelberg cannot be all things to all printers. Its offering in terms of digital printing is rudimentary, for example, and there is no sign of a continuous feed inkjet press with Heidelberg’s name on the side.
The future printer may need such equipment alongside the more traditional highly automated sheetfed litho press, alongside the folding carton line and web portal. All need to be connected, if not to each other, then to a central brain for a print business that is alive to what the sensors, the nerves of the print organism, are feeding back.
As shorter runs and faster turnarounds become the new normal, there simply is not the time for wandering around, picking up job tickets and for production planning meetings to work out what is happening today, tomorrow and for the rest of the week. Everyone knows that a production schedule today is a best guess about what might happen six hours later. Controlling the production schedule, as the printing industry speeds up and has to become even more responsive and responsible to its people and to the environment, will become even more like trying to play four-dimensional chess.
This industry needs connectivity, it needs data and it needs artificial intelligence to shape the decision that shape the success of a business.
Time to connect this industry
Artificial intelligence has become a mainstream topic of conversation, thanks mainly to concerns about text generation software being able to create student essays indistinguishable from those written by two-thumbed students.
This is the Large Language Model flavour of AI where hugely powerful computers and algorithms have been able to parse everything written and published on the internet, have learned the probability of one word following another according to the context in which it is being used, and can construct sentences that follow a logical structure.
The result is that the software can deliver an academic essay, a proposal, an email, thank you note and as part of the campaign to demonstrate the power of the applications, what passes as a poem. And all in seconds and with almost no effort.
ChatGPT has garnered much of the attention, but other LLMs, as they say, are available and many more will be, some better and many worse than those released by Google or Microsoft.
They may not, as it turns out, be winning academic qualifications for students, but they will be creating reports, writing instruction manuals and producing short articles for newspapers, mundane formulaic writing that is a chore to produce. Do not expect War & Peace, though there will certainly be books written entirely by AI. They will not be literary masterpieces or original in any way – and in all probability will never be printed for sale.
There is a key limitation to this type of AI: it draws upon what has already been published. The software is incapable of generating anything that is truly new, of making that unexpected connection that is a hallmark of true creativity. In many ways these algorithms are the next generation of search tools rather than anything more.
This is also just one, highly visible, application of artificial intelligence and one of limited direct use to printers. AI is not just about rearranging words, some are image generating programs that have been gaining attention for mixing bizarre instructions: my grandson playing in the sand in the style of Van Gogh for example – and coming up with equally bizarre images, some of which include hands with six or more fingers. One algorithm was asked to create a picture of salmon swimming up stream and did so, but using images of filleted salmon from the fishmonger’s slab rather than actual fish.
Voice AI is creating some anxiety for it will be possible to use samples of someone’s voice and simulate a conversation that might be used for fraudulent purposes. If it exists online, any content is fair game for manipulation by this technology.
Other AI algorithms are already in place. Certainly AI is already driving chatbots for low level customer support purposes. It is sure to be used in training operators to run pieces of equipment, responding to questions as they arise.
AI is also the next step forward for machine operation. Many press producers are working on applications of AI to automate makeready, sequencing process steps in the right order with no operator intervention, for maintenance where sensors are recording the status of all manner of settings throughout the press and using pattern recognition to understand when something needs attention or is coming close to the end of its life and anticipating component failure and unexpected breakdowns.
Like all AI, this demands vast amounts of data. Indeed the one defining characteristic of any AI application is the hinterland of data that comes before it. AI cannot exist without billions of data points. And any AI that tries to work with a limited data set will only be of limited or broad use. Heidelberg with thousands of presses, each printing dozens of jobs a day, each with hundreds of sensors sending readings back to a server somewhere in the cloud, in real time throughout the day, can establish the patterns that identify upcoming maintenance issues, trouble with a particular ink, the optimum speed and weights for any widely used substrate.
A company with only a handful of users and where machinery is infrequently used will not be able to do this. The depth of data is simply not there. For these companies to share in the spoils of AI, data has to be gathered and in such a way as to impose no additional actions on the company using the equipment that is being harvested for this data. On board sensors are the way to achieve this.
At the moment most of the benefit of AI accrues to the developers and suppliers. They get insight into how machines are being used in the field; they get to identify any weakness that a next iteration might iron out, and they get to manage maintenance call outs by being able to predict and schedule when maintenance is necessary.
Another instance of AI is in automatically generating simple design jobs. Many online design tools are able to create business cards, flyers, menus and more calling upon a vast database of previous designs and can filter these by plumbers, corner cafes, window cleaners and so on, to help shape designs for inexperienced or occasional print buyers. It is not design as a designer would recognise it.
Other applications are able to scale a known label or carton design to create a version specific to a new customer, who is also unlikely to be an experienced designer. Such tools are a way to cope with the deluge of demand for short run packaging that is expected as digital printing becomes established in this sector. For the moment real experience and guidance will win out over the chatbot and parametric design as it is known. These customers need confidence that human communication brings. But this may change in future.
The real gain from AI comes when these designs, whether generated by algorithms or not, reach the production process. Today’s crop of MIS providers can make a reasonable stab at scheduling, but it is limited: all jobs on this weight and type of paper; all jobs to be delivered tomorrow; all jobs needing to be folded into 16pp sections, for example.
They are not always so good at reacting to late changes, a job which is running late, a missing file or lack of paper perhaps. Because this data is lacking. The system can be interrogated after the event, when compiling the invoice say, but live data from every point in the business is vanishingly rare.
Where, for example, is information about a pallet of work in progress? It may be known that the sheets have been printed, with a barcode to help download the data to set up the folder with no further intervention, but until that barcode is scanned, the work in progress has dropped off the radar.
Press suppliers, whether analogue or digital, have concentrated on their piece of equipment not on linking up disparate devices from numerous suppliers.
Connectivity though is slowly coming, whether through the evolution of JDF, APIs or platforms like Zaikio’s Mission Control. Some say the biggest problem with this is its close relationship with Heidelberg and that company’s history of trying to keep its customers within Heidelberg’s walled garden. Hopefully that attitude no longer prevails. Even the mighty Heidelberg must recognise that it cannot dominate every aspect of the printing industry as it might have done in the past. It needs to be open, transparent and to be just as connected as its customers are to their customers.
Despite the changed circumstances, Heidelberg is still in a pivotal position and it believes that its Push to Stop and other software are key differentiating factors that drive machinery sales and keep its factory turning. This is true. But Heidelberg cannot be all things to all printers. Its offering in terms of digital printing is rudimentary, for example, and there is no sign of a continuous feed inkjet press with Heidelberg’s name on the side.
The future printer may need such equipment alongside the more traditional highly automated sheetfed litho press, alongside the folding carton line and web portal. All need to be connected, if not to each other, then to a central brain for a print business that is alive to what the sensors, the nerves of the print organism, are feeding back.
As shorter runs and faster turnarounds become the new normal, there simply is not the time for wandering around, picking up job tickets and for production planning meetings to work out what is happening today, tomorrow and for the rest of the week. Everyone knows that a production schedule today is a best guess about what might happen six hours later. Controlling the production schedule, as the printing industry speeds up and has to become even more responsive and responsible to its people and to the environment, will become even more like trying to play four-dimensional chess.
This industry needs connectivity, it needs data and it needs artificial intelligence to shape the decision that shape the success of a business.
Explore more...
Prayers for the future of the print industry