Sending out street cleaning teams in response to grafitti or rubbish that has been identified using street and mobile cameras rather than simply deploying in regular sweeps. This ensures that teams respond more swiftly to emerging problems. A similar approach can be applied to tending street vegetation, especially trees.

Widely deployed video technology, combined with passport databases, to enable instant, remote passenger recognition to potebntially autoamte the check-in, security and immgration processes. Other management benefits will include warning (and chasing) individuals who are about to miss their flights.

Optimise and target labour allocation for publicly provided services to match demand, especially in areas of constrained supply (e.g. specialist skills)

E-records in health systems are a growing set of critically important documents. Machien learning enables them to be created more efficiently and then updated and (where necessary) consolidated more effectively and cheaply.

Foster Care Technologies use AI to find better matches for children placed in foster care increasing stability in their lives and chances of permanent adoption

Location and establishments hazardous to consumers and individuals might include restaurants, illegal businesses, areas prone to crime and violence and others. Hazards may be medical, environmental, human-caused or other natural phenomena.

Personalise public services to target individual citizens based on multiple data sources (mobile, social media, location, etc). Building up long term data sets on individuals is a growing trend - and one that raises potentially signficant issues alongside the opportunities presented.

Analysing conversation to find words or word combinations that predict likely need for urgent interventions in mental health situations - for example, when suicide risk is at a heightened level. This triage support works best in conjunction with trained human personnel.

Optimise pricing for government provided goods and services (e.g. road pricing, park or museum entrance fees)

Analysing and predicting the trajectory of disease propogation in a population is critical to allow better mobilisation and targeting of scarce resources to combat the spread in a time-sensitive and high pressure situation. Using machine learning to support the process enables modelled predictions to be speedier and less resource-intensive - which can also lead to greater precision and specificity in outcomes.

Censor user generated content on social media and other platforms. or This may be used to reduce hate speech or incitements to violence similar negative issues - but can also be used for political censorship. The line between the two can become a matter of debate.

Prediction modelling typically used to support key decision-making steps in the judicial system - e.g. sentencing and early release. Historically rules and expert based systems there remains perceived risk of bias (e.g. on ethnic grounds) and lack of transparency and accountability.

Mimicing animal behaviour in drones or robots has several potential uses: camouflage (for example in a security situtation), research or media support (allowing humans to better interact with creatures in their natural environment) or replicating (or even replacing) tasks currently performed by natural creatures (pollination for example).

Using wifi signals to identify which individuals are moving - even through physical barriers such as walls. Potential real world uses will include security or facilities tasks - and in the medium term various healthcare or interactive gaming scenarios.

Gathering evidence for criminal prosecution increasingly requries accessing suspect's phones or computers to seek out potential clues - including from photos. In the era of the selfie and cheap computer memory this can potentially involve thousands of images. Using machine learning to identify weapons, drugs or nudity in stored images can considerably speed up the process - although false positives remain an issue.

Using image and location tracking data to identify suspicious behaviour patterns by nautical vessels that might indicate, for example, over-fishing. This would also include multi-vessel activity to predict load transfer. Similar technological approaches might be used to identify other nautical malfeasance such as smuggling.

Identification of "persons of interest" for law enforcement to intercept or track through facial recognition is a high profile use case for AI. Inevitably the debate about the trade-off between individual liberty and security effectiveness plays out differently according to local political expectations. In addition, actual levels of accuracy are a concern both in terms of effective implementation but also civil liberties (algorithmic bias is especially concerning on different ethnic faces for example). In some countries such as China emphasising the supposed effectiveness of the algorithms is a key part of national security communications strategy.

Detect manipulated or falsified media to discover attempts to damage reputations or create unwarranted legal situations. The rise in concerns over `fake news` deployment - especially at such sensitive times as during an election - is a growing area of security and media concern. with potentially huge ramifications.

The rise of biometric-based security systems poses risks that fake credentials - photographs, scans, copies or even post-mortem presentation - may be offered to fool security systems. Image recognition is used to scan for and observe minute differences that may suggest the credentials are invalid.

Identify security-related targets from analysis of sensor feeds (e.g. drone cameras) typically matched with image recognition technology. Increasing debate about the extent to which this is tied to lethal capability and the role of a trained human in the decision loop.

Predict risk of illicit activity or terrorism using historical crime data, intelligence data and other available sources (e.g. predictive policing). Largely based on tracking network traffic, most obviously social media and telecoms activity, to predict likelihood and risk of potential terror-related criminal activity. The techniques will also reveal potential suspects. Similar techniques may be used to track other "undesirables" - including pro-democracy protesters in more repressive regimes.

Analyse data feeds about incoming cyber activity as well as behavioural data inside an organisation - married with up to date information about the broader cyber security situation - to predict potential security risks.

Growing field of military vehicle automation. A key question is the degree of autonomous decision-making on fire / no-fire lethal decisions. The key benefit of reduced risk to military personnel is not likely to be equally true of opponents. Growing legal and moral questions likely an area of rapidly diverging decision-making across nation states and, potentially, non-state actors.

A key part of hacking or similar scams such as phishing is building a larger database of potential targets and a better understanding of their profiles. One AI technique to support this is to use facial recognition across platforms to identify multiple accounts. Depending on regulatory data rights similar techniques can be used for more conventional marketing.

Use of video footage potentially combined with "lie detector" analysis to make preliminary screening decisions on travellers entering protected borders. The aim is both to speed up the process and - more controversially - to achieve greater accuracy in screening results.

Espionage related cracking of opponents secure communications. Techniques used by many actors - but states still retain critical edge.

Deployed military installations will increasingly use AI for autonomous decision-making - often with a lethal impact. Examples include sentry guns (deployed on the Korean border) and so-called 'intelligent mines'.

Determine whether there is attempted and likely cyber breaches that could impact sensitive data and company operations. Also follow compliance reporting regulations for breaches.

Volumes of mail are such that it is almost impossible to consistently check for items of concern - whether smuggled drugs or packages of suspicious substances likes anthrax. Using X-rays or similar sensor technology large volumes of images can be processed to indicate which items of mail should be properly investigated.

Utilise learning systems to effectivel and swiftly respond to security threats, many of which may have been delivered with machine learning support. This is a game where every innovation in defence triggers the next innovation in attack - and vice versa.