AI = logic and discipline

Artificial Intelligence (AI) is commonly defined as systems and algorithms capable of performing tasks that would normally require human intelligence, typically in relation to:

• Signal recording, e.g. still images, live recordings, sound, speech, text, pulse, temperature, alkalinity, GPS, and network data, pattern recognition, for example, qualities, defects, breaks, tumors, faces, and speech;
• Integration of data into information, and integration of information into meanings, to use the result for a given purpose, optimised to some definition of success, for example, for reasoning, problem solving, decision making and adaptation to changed conditions and new situations.

In short, AI is any system able to integrate data, recognise patterns, and use the result purposefully. The relationship can be succinctly expressed as Data + Task = Application, as a means to illustrate the broad potential of AI.

AI encompasses a variety of disciplines. Large Language Models (LLMs) like ChatGPT and Gemini represent a subset of Generative AI, which in turn is a branch of Deep Learning. Deep Learning is built upon Neural Networks, a type of Machine Learning, all are components of AI.

The goal for the development of AI is to create systems that are self-learning across domains, without specific programming for each task and without the need for human intervention. This requires AI to be able to extract categorical properties of exemplars, a defining feature of adaptive organisms in evolutionary systems.

Red frogs

If a living organism, it is able to put data together, e.g. red + frog, it creates the information that there are red frogs. If it puts information together, e.g. red frog + tastes bad, it creates the knowledge that red frogs taste bad. If it puts the this knowledge into context, e.g. red frog + tastes bad + you die the next day, it creates a crucial evolutionary insight. An organism that is able to extract categorical properties from exemplars, e.g. frog shape and red colour, and attach strong emotions to red frog encounters, has a much better chance of survival.

In production, this means that if you are able to recognise even subtle relationships between operational characteristics and production outcomes, e.g. too high a temperature, and attach warning systems to them, you have a much better chance of surviving in a complex production setting.

The middle layer

In the first, basic layer of tangible assets, engineers and craftsmen set up machines and workstations. In the third layer of dynamic execution, operators operate the machines and workstations, and churn out products and services. In between these two, there is a less tangible layer of data, information, and knowledge, which makes the dynamics possible. With larger and more mechanised productions, the amount of data increases, and with digitisation of machines, processes, and signal recording, this middle data layer has become even larger and more comprehensive, and also more decisive and accessible.

Formalised systems and software have made it possible to move knowledge and heuristics (mental shortcuts that allow people to make fast decisions but which can also lead to cognitive bias) out of the heads of skilled people and into manuals, bills-of-materials, quality systems, ERP systems, and so on. AI is radically going to change this, because AI can be developed into kinds of autopilots that choose intelligently and self-manage processes towards predefined goals, independent of human intervention.We see this with self-driving cars and the thrilling, but somewhat eerie, New Atlas robots from Boston Dynamics as an ultimate result of embodied AI.

Old news

We have had AI for decades and have met it in a number of forms and guises. In its simplest form AI is algorithms, for instance for linear fit of data that can predict developments and directions, or mathematical object functions that can find extremes in linear data, e.g. cutting metal shapes with a laser. Advanced twodimensional cutting optimisation programs were available in the 1980s, so ChatGPT is perhaps just the first time many of us have been aware that we were facing AI as such.

Here are a number of more relevant examples from our own industry.

Data

Data categories you typically find in laundries are many, these are just some of the examples:

• Tabular data: Customers, machines, textiles, cars, items, persons, numbers
• Time series data:
• Historical: Meters (consumptions, lengths, items, hours)
• Current: Sensors (power, temperature, pH, speed), counts (pcs./min.)
• Forecasts: Loads (textiles, person hours), results (ltr./kg., kg./hr., M€/yr.)
• Images: Photos (infrared, x-ray), surface scans (textiles)
• Texts: Internal & external (emails, minutes, forms, chats)
• Network data: User data (websites, campaigns).

Tasks

Information handling tasks in our industry range from decision support to automation. Data examples are:

Recognition:

• Categorisation (bar codes, RFIDS), classification (machine alerts, stains)
• Image recognition (pocket objects), data analysis (batch loads, performance metrics), post-wash inspection (cleanliness)
• Load recognition (false & genuine bottlenecks)

Forecasting:

• Data fitting (usage, failure rates and down time; moisture rentention & drying times; volumes & work station loads)
• Time series prediction (seasonal variations, hotel occupancy rates & laundry volumes)

Controlling:

• Denoising (meter, sensor & monitor data)
• Anomaly detection (deviations in sensor data & work station speeds)

Optimising:

• Sequential data flow (batch sequences, operation sequences, load sequences)
• Feedback loops (post-wash data fed to washers & driers; AI-internal learning feedback)
• Adversarial networks (AI-internal improvement, process simulation)

Designing:

• Simulation (virtual prototyping, proces modeling, time-line scenarios)
• Testing (test automation, stress testing, maintenance prediction)

Generating:

• Natural language processing, speech recognition
• Machine translation, speech synthesis
• Text summarisation, text generation.

Applications

Here comes the short expression Data + Task = Application in handy to illustrate the many and very diverse applications in our own industry. These are just a few examples:

• Time series + forecasting = Demands, consumptions, capacity loads
• Images + categorisation = Textile categorisation, item detection
• Network + optimisation = Bag call-off, process routing, employee allocation
• Time series + controlling = Speeds, temperatures, chemical dosing, predictive alerts
• Images + categorisation = Spot & defect detection, malfunction detection
• Tabular + categorisation = Error & malfunction detection
• Network + optimisation = Route planning, GPS navigation
• Texts + generating = Chatbots, sales letters, images
• Time series + designing = Machines, textiles, laundry layouts.

So…

What should we expect to see in our own industry in the near future? It would be justified to expect:

• Automatic sorting-in and item counting
• Ironer lines that adjust their speed to the batch’s actual residual moisture
• Tunnel washers that dose chemicals according to chemical activity
• Call-off-systems, that autopilot the wash-room through the thousand bags of a day
• Smart bins, that categorise and count items en route to the laundry
• Machine design mimicking nature’s design principles.

In all parts of our laundries we are going to see drill depth, predictive power, and decision support that will increase the number and complexity of variables, but at the same time reduce the number of relevant options. We may even see AI systems solve the hardest problem of them all – making a laundry able to realise the potential in its product mix meeting with its assets, regardless of the manager in charge.