Even with advances in computing power there will come a moment when too much data has been gathered to be economically stored - for example during high spec science experiments. At that stage, in real time, machine learning can be used to help decide which data should be stored for analysis and which deleted.

Synthetic datasets is a way for researchers to train machine learning programs when sufficient real-world data is unavailable, difficult to obtain, or raises ethical or privacy concerns. Certain use cases for AI lend themselves better to synthetic datasets than others.

Students expressions and movements are analysed to check that they are paying attention in class. The system will be able to tell if students are reading or listening – or napping at their desks. Students will get a real-time attentiveness score, which will be shown to their teacher on a screen.

Analyse and grade school or college work. Using machine learning to analyse key elements this will give grades - potentially on a national or international scale compared to individual teacher's marks. This will tend to reward conformity over originality so will be better suited for some courses and topics than others. It could potentially support roll-out of education to areas with limited educational resources (e.g. lack of teaching staff).

Across large text datasets - for example, historical archives - there will be trends or insights that can be analysed but which it would take a human observer too much time to process. Although still at a relatively rudimentary stage this will become an increasingly powerful tool for creating new insights and levels of knowledge.

Predict individualised educational and career paths to maximise engagement and success - both on an economy-wide basis but also for individuals. Ideally this will help inform better life decisions.

Provide personalised learning programmes - with regular testing / feedback loops used to assess and deliver against topics that individual students find more challenging and a pace, and potentially style, of interaction tailored to the individual.

Using algorithms to allocate access to public goods. This may be to distribute access fairly, efficiently, effectively - or for more broad political goals (minimise dissent and reward behaviour).

Monitoring and identifying the illicit use of natural resources for activities such as illegal farming, logging, deforestation, etc.

Merge data flows from multiple sources - visual (satellite, drone, CCTV, Google Street Map etc), demographic and socio-economic, sensors (e.g. traffic flow), social media and other - to build models for prediction and flow modelling and analytics of the city as a system. can be used for short term tactical decisions - traffic stoppages for example - or long term strategic - planning decisions for example. AI co-ordinates, calculates and potentially delivers visualisation.

Forecast macroeconomic variables based on huge volumes of government proprietary and public data - often combined with other data sources to optimise forecasting.

Optimise public policy decisions (e.g., housing) to take into account greater set of complex interactions. AI can be especially useful in modelling some of the potential second or third order effects of decisions but clear analyst question formulation will be key. This requires manipulating multiple data sources to help define decision-making options, and recommend preferential outcomes. This is an area increasingly being investigated by defence and security state-level actors and is therefore often relatively opaque in terms of detailed information.

Replicate (often back-office decision) processes for applications, permitting, tax auditing, etc

Optimise public resource allocation for urban development to improve quality of life (e.g., reduce traffic, minimise pollution). This may be based on multiple input models.

Optimise procurement strategy to reduce costs for large government agencies (e.g., Defence). This is especially viable when used to repeat purchase lower value, commoditised goods - e.g. food or ammunition rather than aircraft carriers.

Using a full spectrum of economic, social, demographic, physical, health and other data to model and predict likely health outcomes in a given population with a focus on emerging chronic and mental health issues. This can help with scenario and investment planning - e..g the impact of housing investment or employment losses on likely mental issues.

Predict risk from natural causes such as wildfires, using data from a wide variety of sources such as satellites and environmental sensors and often building on other data models - for example likely weather patterns. Predicting likely wildfire movement for example can have huge impact on human safety and resource deployment.

Predict migration volume based on different data sources and indicators. This can aid country policymakers and NGOs to prepare for changing migration patterns.

Global distribution and connectivity relieson increasingly accurate mapping to optimise supply chains, distribution and access. Using satellite and other remotely captured images it is increasingly possible to build accurate maps at scale and provide addressing schemes in underserved areas.

Predict likelihood of traffic accidents or queues and optimise traffic system to reduce the risk. This will require the deployment of sensors and 'smart' traffic lights.

Vast amounts of data has been generated by an array of human sensors like telescopes or exploratory spacecraft examining the mysteries of space. Analysing data at this scale is beyond human abilities - especially when part of the challenge is to find things that we did not know existed - so machine learning plays a vital role. Ultimately this may lead to the discovery of something as exciting as extra-terrestrial beings - but that is probably at least as distant as the creation of Artificial General Intelligence....

Improve weather forecasting accuracy using machine learning applied to multiple data sources. The aim is to improve an existing process that is typically already a sophisticated user of computer modelling.

Using social media to compile information to identify service disruptions. Angry or worried tweeting (or similar) by customers or announcements by operators will trigger consolidated public service alerts on service disruptions. This is most advanced in transport networks but could be applied to other services.

Use satellite imagery to identify areas that need the most support and assistance due to lower levels of development or emerging agricultural issues (e.g. poor harvests)

Using cameras to take photos of environmentally sensitive sites - whether above or below the waves - which can then be analysed, and the contents of pictures classified, to estimate their overall health status. Issues that might be included would be bio-diversity of flora and health status of things like coral. Future iterations might assess small creatures - fish or insects - to add depth to the calculations.

Deployment of falsified media content with the aim of influencing public opinion - either by setting up new narratives or undermining existing narratives. "Memes" are typically deployed.

Provide access to digital information to illiterate folk through conversational agent tools / chatbots. Voice interaction enables new forms of communication - for example low / free marginal cost access to healthcare advice or other information. Access to digital devices, Internet and electricity may be factors limiting deployment.

Protect individual's privacy by changing key pixels in photos to confuse AI driven image recognition. Typically these will not be immediately discernible to the human eye if, for example, they are published. The photo might be put through an app to ensure that that is is subtly changed. Inevitably counter software (AI-driven) can be deployed.

Detecting, identifying and monitoring animal populations is especially useful when they are in the wild and potentially at risk. Typically it requires a network of static camera traps or acoustic recorders but can also be extended to include mobile sensors.

Modelling populations using wide variety of demographic, polling, socio-economic and online activity data to predict the likelihood that individuals / segments will be receptive to political (or related) messaging. May model appropriate type, mode and style of messaging as well as optimal influencers / messengers,.