Please note: The algorithm descriptions in English have been automatically translated. Errors may have been introduced in this process. For the original descriptions, go to the Dutch version of the Algorithm Register.
Salesforce Einstein Chatbot
- Publication category
- Impactful algorithms
- Impact assessment
- DPIA, ...
- Status
- In use
General information
Theme
Begin date
Contact information
Link to publication website
Responsible use
Goal and impact
The Amsterdam municipality is engaged in a programme to improve contact with Amsterdammers.
The chatbot contributes to this goal: a new low-threshold automated communication channel where citizens receive predefined answers 24/7 from a chatbot and (within opening hours) can be connected to an employee.
Einstein is a chatbot that can communicate with a resident using natural language recognition. Einstein uses an algorithm to recognise the intent of a question asked.
This algorithm is used to trigger one of the following actions:
- Providing a predefined answer to frequently asked questions.The algorithm does not create answers.
- Transferring the call to the appropriate department.
The algorithm is self-learning because it learns to better recognise residents' intentions. Forwarding is done based on a routing table managed by the Amsterdam municipality. This is not self-learning.
Considerations
Using a chatbot alongside existing communication channels makes the municipality of Amsterdam more accessible. Citizens can choose between different communication channels.
A chatbot meets the need for 24/7 service. It also provides residents with faster answers to questions.
Because common questions are handled by the chatbot, employees have more time for personal contact with citizens.
The chatbot does not ask for personal data, but the user may still share personal data. The chatbot's welcome message therefore starts the conversation with the following warning: Remember your privacy, do not share personal data.
To reduce the risk of wrong answers, we do not use generative AI. All answers are pre-formulated and checked by humans.
Also, no data from the Amsterdam municipality is used to train the NLP algorithm.
Residents who prefer contact with a person can be put through to a staff member (within opening hours).
Human intervention
Training and adjusting the models used by the chatbot is done by humans. As will the monitoring of the chatbot's performance and conversations. Complex questions or unclear outcomes will be referred to a human.
Questions requiring personal data will also always be referred to a human.
Risk management
Vendor Salesforce uses best practices, tools and frameworks for developing and using Salesforce products to ensure that its products, features, models and apps are trustworthy, avoid bias and discriminatory outcomes and prioritise transparency.
More information can be found at
Responsible AI and Technology | Salesforce US
Has also joined the:
EU Artificial Intelligence Act
US AI Safety Institute Consortium
The chatbot does not ask for personal data, but the user may still share personal data. The chatbot's welcome message therefore starts the conversation with the following warning: Remember your privacy, do not share personal data.
To reduce the risk of wrong answers, we do not use generative AI. All answers are pre-formulated and checked by humans.
Also, no data from the Amsterdam municipality is used to train the NLP algorithm.
Residents who prefer to contact a person can be connected (within opening hours) to a staff member.
Legal basis
The General Data Protection Regulation (GDPR) defines 'tasks in the public interest' as tasks related to the exercise of public authority or tasks carried out by a body entrusted with a public task. Services provided by the municipality fall within this definition.
Link to Processing Index
Elaboration on impact assessments
Ethics leaflet implemented; no IAMA required.
Impact assessment
- Data Protection Impact Assessment (DPIA)
- Ethische bijsluiter Amsterdam
Operations
Data
Salesforce does not use external data to train its NLP models, but uses data from its customers.
The municipality of Amsterdam has opted out. As a result, data from Amsterdam is not used to train algorithms.
Technical design
The chatbot uses Natural Language Processing (NLP) to recognise the intent of the question from free text (from the chat conversation).