Artificial Intelligence Policy

The purpose of this document is to outline the approach taken by Commonplace Digital Ltd to artificial intelligence (AI) in its business processes and platform.

Artificial intelligence is described as technology that enables computers and machines to simulate human learning, comprehension, problem solving, decision making, creativity and autonomy. 

Commonplace collects data from respondents on behalf of its customers and partners (‘Engagement Partners’) in line with its privacy policy. 

Artificial Intelligence is used in the following functions, described in greater detail below:

• Data analysis (TrendsAI)
• Voice capture
• Language translation

Commonplace also relies on a number of other algorithms to power our systems and processes such as anti-gaming, moderation and journey step recommendations. These do not use AI; instead they are powered by rules defined by the Commonplace team. 

TrendsAI

The Commonplace TrendsAI product relies on machine learning for a number of activities under the remit of Natural Language Processing (NLP) which describes the ability of a computer program to understand human language as it's spoken and written:

• Text classification - assigning topics to free text comments
• Sentiment classification - assigning sentiment and/or emotion scores to free text comments
• Summarisation of data - generating written summaries to generate insights

All three processes are reliant on the use of NLP and the second / third items also rely on Large Language Models (LLM), which are machine-learning models powered by neural networks. They learn to understand statistical relationships from large amounts of text data. 

Where our Engagement Partners have opted into the TrendsAI feature, the software will work to deliver the enhanced analysis to their project dashboard. TrendsAI is not a solution that Commonplace has procured off-the-shelf; it is custom built by Commonplace for our and our customers’  specific needs. The NLP models rely on external global and publicly available data to support them in their evolution. The accuracy of our NLP models is additionally reliant on fully anonymised data collected by Commonplace over the course of its activities, which has been used to train and tune the models to the specific domain of the built environment. At no point has Commonplace data been stored by any open source NLP models or used to train or enhance the accuracy of their models. Commonplace data is only made available to our proprietary version of the NLP models. 

Commonplace has invested significantly, not only developing, but also in testing our NLP models, with a particular focus on accuracy, precision and reliability. Our regular and ongoing testing provides a high level of confidence that our NLP models will correctly predict topics and sentiments, and generate summaries to a high level of accuracy. The results of the NLP models are available to our Engagement Partner users in their Commonplace dashboard, which allows them to review the accuracy and relevance of the results. There may be occasions when our users do not agree with the results presented by Trends AI, and as such, we present them with the option to flag poor results or to build their own manual classification rules. 

Where TrendsAI is being used to perform statutory  processes, such as evaluation of planning application comments or local plan representations, it is important to be aware of the role AI has to play, particularly in the context of the  PINS guidance on use of AI in the planning system. Although the PINS guidance does not apply to the majority of Commonplaces (which are not subject to planning appeals, development consent orders, examinations of local plans and other planning-related and specialist casework), we anticipate that this policy will become the industry standard for decisions of a planning nature.

If you plan to use TrendsAI outputs in your evaluation or presentation of results, we’d recommend that you always: 

• Clearly label where you have used AI and whether you have used it to 
  • Classify topics from free text answers
  • Classify sentiment / emotion from free text answers
  • Summarise data in order to generate insights
• Describe why you used AI and how it helped you to achieve your goals
• Explain that you have reviewed the outputs of TrendsAI and that you are satisfied with its factual accuracy.
• Emphasise that 
  • Survey answers are processed by external AI models
  • Survey answers are not stored by external AI models, nor used to train them
  • Demographics and other personally identifiable information is not processed by external AI models

Commonplace can declare that the use of Trends AI is responsible and lawful and that the system actively works to identify and mitigate any bias to ensure fairness, inclusivity and not to discriminate against any individuals or groups.

Here is an example AI statement that you may wish to include in your reports:

This report is based on data collected via the Commonplace platform, which was processed using their TrendsAI solution, in accordance with the Commonplace AI policy. 

We used TrendsAI for the free text question ‘What other suggestions do you have for this area’ question and TrendsAI processed the analysis on 3 Oct 2024.

We used TrendsAI to support our evaluation of the project by classifying free text answers so that we can understand the key topics and the sentiment of our respondent community towards each topic. Without Commonplace’s tools, we would have been prohibited to carry out detailed analysis on free text answers due to the cost of manual analysis. In addition, given the accuracy of Commonplace’s continuously tested feature, we also feel that it generates more reliable and usable insights than a team of people who may add subjectivity. 

In addition, we used TrendsAI’s summarisation feature to support us in the generation of the topic summaries you will find in this report. Our team reviewed each summary produced by the Commonplace and edited it according to our needs and tone-of-voice guidelines, while at the same time reviewing the accuracy of the summaries.

We have reviewed the outputs of TrendsAI and are satisfied with its factual accuracy.

Commonplace takes data protection seriously, such that your answers are not used to train any external AI models and your personal information is not passed to any external service for the purpose of AI. 

Commonplace declares that the use of Trends AI is responsible and lawful and that the system actively works to identify and mitigate any bias to ensure fairness, inclusivity and not to discriminate against any individuals or groups.

It is important for our customers to understand potential limitations associated with the use of NLP and LLMs. While TrendsAI’s development has worked hard to mitigate these limitations, it is not possible to rule them out. The TrendsAI product encourages human spot-checking with the ability to correct the automated output, and the LLM learns from these human interactions. It is also important to note that even with these limitations accounted for, our research shows that TrendsAI is more accurate than a human doing the same task. 

• Edge cases refer to unusual, rare, or exceptional situations that aren't well represented in the training data that we used to develop the models. These cases can lead to limitations in the outputs, such as model overconfidence, misinterpretation of context, or inappropriate outputs.

• Model hallucinations, grounding, and factuality in real-world knowledge may result in outputs that are plausible-sounding but factually incorrect, irrelevant, inappropriate or nonsensical. 

• Language models can inadvertently amplify existing biases in their training data, leading to outputs that might further reinforce societal prejudices and unequal treatment of certain groups.

• Text generation has been optimised for British English and may not be as effective for some dialects or language varieties because they are underrepresented in the training data.

Voice capture

For customers wishing to take advantage of our voice capture question, AI is used in order to transcribe the voice messages left by respondents. This service is provided by a third party, more information about this service can be found here: https://www.speeqo.com/privacy

Content standards

For customers wishing to take advantage of translation of the platform static content or dynamic content that they publish on the platform, AI may be used for machine translation of the content. This service is provided by a third party, more information about this service can be found here: https://www.smartcat.com/machine-translation  

This policy is effective 14 Oct 2024 and was last updated on 14 Oct 2024. If you have any questions or concerns, you may contact us at privacy@commonplace.is